1 2018-01-26 Jakub Jelinek <jakub@redhat.com>
3 PR rtl-optimization/83985
4 * dce.c (deletable_insn_p): Return false for separate shrink wrapping
6 (delete_unmarked_insns): Don't ignore separate shrink wrapping
7 REG_CFA_RESTORE insns here.
10 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Don't
11 use SSA_NAME_VAR as base for SSA_NAMEs with non-NULL SSA_NAME_VAR.
13 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
15 * config/arc/arc-arch.h (arc_tune_attr): Add ARC_TUNE_CORE_3.
16 * config/arc/arc.c (arc_sched_issue_rate): Use ARC_TUNE_... .
18 (arc_override_options): Likewise.
19 (arc_file_start): Choose Tag_ARC_CPU_variation based on arc_tune
21 (hwloop_fail): Use TARGET_DBNZ when we want to check for dbnz insn
23 * config/arc/arc.h (TARGET_DBNZ): Define.
24 * config/arc/arc.md (attr tune): Add core_3, use ARC_TUNE_... to
25 properly set the tune attribute.
26 (dbnz): Use TARGET_DBNZ guard.
27 * config/arc/arc.opt (mtune): Add core3 option.
29 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
31 * config/arc/arc.c (arc_delegitimize_address_0): Refactored to
32 recognize new pic like addresses.
33 (arc_delegitimize_address): Clean up.
35 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
37 * config/arc/arc-arches.def: Option mrf16 valid for all
39 * config/arc/arc-c.def (__ARC_RF16__): New predefined macro.
40 * config/arc/arc-cpus.def (em_mini): New cpu with rf16 on.
41 * config/arc/arc-options.def (FL_RF16): Add mrf16 option.
42 * config/arc/arc-tables.opt: Regenerate.
43 * config/arc/arc.c (arc_conditional_register_usage): Handle
44 reduced register file case.
45 (arc_file_start): Set must have build attributes.
46 * config/arc/arc.h (MAX_ARC_PARM_REGS): Conditional define using
48 * config/arc/arc.opt (mrf16): Add new option.
49 * config/arc/elf.h (ATTRIBUTE_PCS): Define.
50 * config/arc/genmultilib.awk: Handle new mrf16 option.
51 * config/arc/linux.h (ATTRIBUTE_PCS): Define.
52 * config/arc/t-multilib: Regenerate.
53 * doc/invoke.texi (ARC Options): Document mrf16 option.
55 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
57 * config/arc/arc-protos.h: Add arc_is_secure_call_p proto.
58 * config/arc/arc.c (arc_handle_secure_attribute): New function.
59 (arc_attribute_table): Add 'secure_call' attribute.
60 (arc_print_operand): Print secure call operand.
61 (arc_function_ok_for_sibcall): Don't optimize tail calls when
63 (arc_is_secure_call_p): New function. * config/arc/arc.md
64 (call_i): Add support for sjli instruction.
65 (call_value_i): Likewise.
66 * config/arc/constraints.md (Csc): New constraint.
68 2018-01-26 Claudiu Zissulescu <claziss@synopsys.com>
69 John Eric Martin <John.Martin@emmicro-us.com>
71 * config/arc/arc-protos.h: Add arc_is_jli_call_p proto.
72 * config/arc/arc.c (_arc_jli_section): New struct.
73 (arc_jli_section): New type.
74 (rc_jli_sections): New static variable.
75 (arc_handle_jli_attribute): New function.
76 (arc_attribute_table): Add jli_always and jli_fixed attribute.
77 (arc_file_end): New function.
78 (TARGET_ASM_FILE_END): Define.
79 (arc_print_operand): Reuse 'S' letter for JLI output instruction.
80 (arc_add_jli_section): New function.
81 (jli_call_scan): Likewise.
82 (arc_reorg): Call jli_call_scan.
83 (arc_output_addsi): Remove 'S' from printing asm operand.
84 (arc_is_jli_call_p): New function.
85 * config/arc/arc.md (movqi_insn): Remove 'S' from printing asm
87 (movhi_insn): Likewise.
88 (movsi_insn): Likewise.
89 (movsi_set_cc_insn): Likewise.
90 (loadqi_update): Likewise.
91 (load_zeroextendqisi_update): Likewise.
92 (load_signextendqisi_update): Likewise.
93 (loadhi_update): Likewise.
94 (load_zeroextendhisi_update): Likewise.
95 (load_signextendhisi_update): Likewise.
96 (loadsi_update): Likewise.
97 (loadsf_update): Likewise.
98 (movsicc_insn): Likewise.
99 (bset_insn): Likewise.
100 (bxor_insn): Likewise.
101 (bclr_insn): Likewise.
102 (bmsk_insn): Likewise.
103 (bicsi3_insn): Likewise.
104 (cmpsi_cc_c_insn): Likewise.
105 (movsi_ne): Likewise.
106 (movsi_cond_exec): Likewise.
107 (clrsbsi2): Likewise.
118 (call_i): Remove 'S' asm letter, add jli instruction.
119 (call_value_i): Likewise.
120 * config/arc/arc.op (mjli-always): New option.
121 * config/arc/constraints.md (Cji): New constraint.
122 * config/arc/fpx.md (addsf3_fpx): Remove 'S' from printing asm
124 (subsf3_fpx): Likewise.
125 (mulsf3_fpx): Likewise.
126 * config/arc/simdext.md (vendrec_insn): Remove 'S' from printing
128 * doc/extend.texi (ARC): Document 'jli-always' and 'jli-fixed'
130 * doc/invoke.texi (ARC): Document mjli-always option.
132 2018-01-26 Sebastian Perta <sebastian.perta@renesas.com>
134 * config/rl78/rl78.c: if operand 2 is const avoid addition with 0
135 and use incw and decw where possible
136 * testsuite/gcc.target/rl78/test_addsi3_internal.c: new file
138 2018-01-26 Richard Biener <rguenther@suse.de>
140 PR tree-optimization/81082
141 * fold-const.c (fold_plusminus_mult_expr): Do not perform the
142 association if it requires casting to unsigned.
143 * match.pd ((A * C) +- (B * C) -> (A+-B)): New patterns derived
144 from fold_plusminus_mult_expr to catch important cases late when
145 range info is available.
147 2018-01-26 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
149 * config/i386/sol2.h (USE_HIDDEN_LINKONCE): Remove.
150 * configure.ac (hidden_linkonce): New test.
151 * configure: Regenerate.
152 * config.in: Regenerate.
154 2018-01-26 Julia Koval <julia.koval@intel.com>
156 * config/i386/avx512bitalgintrin.h (_mm512_bitshuffle_epi64_mask,
157 _mm512_mask_bitshuffle_epi64_mask, _mm256_bitshuffle_epi64_mask,
158 _mm256_mask_bitshuffle_epi64_mask, _mm_bitshuffle_epi64_mask,
159 _mm_mask_bitshuffle_epi64_mask): Fix type.
160 * config/i386/i386-builtin-types.def (UHI_FTYPE_V2DI_V2DI_UHI,
161 USI_FTYPE_V4DI_V4DI_USI): Remove.
162 * config/i386/i386-builtin.def (__builtin_ia32_vpshufbitqmb512_mask,
163 __builtin_ia32_vpshufbitqmb256_mask,
164 __builtin_ia32_vpshufbitqmb128_mask): Fix types.
165 * config/i386/i386.c (ix86_expand_args_builtin): Remove old types.
166 * config/i386/sse.md (VI1_AVX512VLBW): Change types.
168 2018-01-26 Alan Modra <amodra@gmail.com>
171 * config/rs6000/rs6000-p8swap.c (rtx_is_swappable_p): Exclude
172 UNSPEC_VBPERMQ. Sort other unspecs.
174 2018-01-25 David Edelsohn <dje.gcc@gmail.com>
176 * doc/invoke.texi (PowerPC Options): Document 'native' cpu type.
178 2018-01-25 Jan Hubicka <hubicka@ucw.cz>
181 * predict.c (drop_profile): Do not push/pop cfun; update also
183 (handle_missing_profiles): Fix logic looking for zero profiles.
185 2018-01-25 Jakub Jelinek <jakub@redhat.com>
188 * ipa-fnsummary.c (compute_fn_summary): Clear can_change_signature
189 on functions with #pragma omp declare simd or functions with simd
191 * omp-simd-clone.c (expand_simd_clones): Revert 2018-01-24 change.
192 * config/i386/i386.c (ix86_simd_clone_compute_vecsize_and_simdlen):
193 Remove trailing \n from warning_at calls.
195 2018-01-25 Tom de Vries <tom@codesourcery.com>
198 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
199 for neutered workers.
201 2018-01-24 Joseph Myers <joseph@codesourcery.com>
204 * config/m68k/m68k.c (m68k_promote_function_mode): New function.
205 (TARGET_PROMOTE_FUNCTION_MODE): New macro.
207 2017-01-08 Jeff Law <law@redhat.com>
210 * i386.c (get_probe_interval): Move to earlier point.
211 (ix86_compute_frame_layout): If -fstack-clash-protection and
212 the frame is larger than the probe interval, then use pushes
213 to save registers rather than reg->mem moves.
214 (ix86_expand_prologue): Remove conditional for int_registers_saved
217 2018-01-24 Vladimir Makarov <vmakarov@redhat.com>
220 * ira-build.c (setup_min_max_allocno_live_range_point): Set up
221 min/max for never referenced object.
223 2018-01-24 Jakub Jelinek <jakub@redhat.com>
226 * tree.c (free_lang_data_in_decl): Don't clear DECL_ABSTRACT_ORIGIN
228 * omp-low.c (create_omp_child_function): Remove "omp declare simd"
229 attributes from DECL_ATTRIBUTES (decl) without affecting
230 DECL_ATTRIBUTES (current_function_decl).
231 * omp-simd-clone.c (expand_simd_clones): Ignore DECL_ARTIFICIAL
232 functions with non-NULL DECL_ABSTRACT_ORIGIN.
234 2018-01-24 Richard Sandiford <richard.sandiford@linaro.org>
236 PR tree-optimization/83979
237 * fold-const.c (fold_comparison): Use constant_boolean_node
238 instead of boolean_{true,false}_node.
240 2018-01-24 Jan Hubicka <hubicka@ucw.cz>
242 * ipa-profile.c (ipa_propagate_frequency_1): Fix logic skipping calls
245 2018-01-24 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
247 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
248 Simplify the clause that sets the length attribute.
249 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
250 (*sibcall_nonlocal_sysv<mode>): Clean up code block; simplify the
251 clause that sets the length attribute.
252 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
254 2018-01-24 Tom de Vries <tom@codesourcery.com>
257 * config/nvptx/nvptx.c (WORKAROUND_PTXJIT_BUG_2): Define to 1.
258 (nvptx_pc_set, nvptx_condjump_label): New function. Copy from jump.c.
259 Add strict parameter.
260 (prevent_branch_around_nothing): Insert dummy insn between branch to
261 label and label with no ptx insn inbetween.
262 * config/nvptx/nvptx.md (define_insn "fake_nop"): New insn.
264 2018-01-24 Tom de Vries <tom@codesourcery.com>
267 * config/nvptx/nvptx.c (nvptx_single): Add exit insn after noreturn call
268 for neutered threads in warp.
269 * config/nvptx/nvptx.md (define_insn "exit"): New insn.
271 2018-01-24 Richard Biener <rguenther@suse.de>
273 PR tree-optimization/83176
274 * tree-chrec.c (chrec_fold_plus_1): Handle (signed T){(T) .. }
277 2018-01-24 Richard Biener <rguenther@suse.de>
279 PR tree-optimization/82819
280 * graphite-isl-ast-to-gimple.c (binary_op_to_tree): Avoid
281 code generating pluses that are no-ops in the target precision.
283 2018-01-24 Richard Biener <rguenther@suse.de>
286 * tree-cfg.c (replace_loop_annotate): Handle annot_expr_parallel_kind.
288 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
290 * cfgcleanup.c (try_crossjump_to_edge): Use combine_with_count
291 to merge probabilities.
292 * predict.c (probably_never_executed): Also mark as cold functions
293 with global 0 profile and guessed local profile.
294 * profile-count.c (profile_probability::combine_with_count): New
296 * profile-count.h (profile_probability::operator*,
297 profile_probability::operator*=, profile_probability::operator/,
298 profile_probability::operator/=): Reduce precision to adjusted
299 and set value to guessed on contradictory divisions.
300 (profile_probability::combine_with_freq): Remove.
301 (profile_probability::combine_wiht_count): Declare.
302 (profile_count::force_nonzero):: Set to adjusted.
303 (profile_count::probability_in):: Set quality to adjusted.
304 * tree-ssa-tail-merge.c (replace_block_by): Use
307 2018-01-23 Andrew Waterman <andrew@sifive.com>
308 Jim Wilson <jimw@sifive.com>
310 * config/riscv/riscv.c (riscv_stack_boundary): New.
311 (riscv_option_override): Set riscv_stack_boundary. Handle
312 riscv_preferred_stack_boundary_arg.
313 * config/riscv/riscv.h (MIN_STACK_BOUNDARY, ABI_STACK_BOUNDARY): New.
314 (BIGGEST_ALIGNMENT): Set to STACK_BOUNDARY.
315 (STACK_BOUNDARY): Set to riscv_stack_boundary.
316 (RISCV_STACK_ALIGN): Use STACK_BOUNDARY.
317 * config/riscv/riscv.opt (mpreferred-stack-boundary): New.
318 * doc/invoke.tex (RISC-V Options): Add -mpreferred-stack-boundary.
320 2018-01-23 H.J. Lu <hongjiu.lu@intel.com>
323 * config/i386/i386.c (ix86_expand_prologue): Use cost reference
324 of struct ix86_frame.
325 (ix86_expand_epilogue): Likewise. Add a local variable for
326 the reg_save_offset field in struct ix86_frame.
328 2018-01-23 Bin Cheng <bin.cheng@arm.com>
330 PR tree-optimization/82604
331 * tree-loop-distribution.c (enum partition_kind): New enum item
332 PKIND_PARTIAL_MEMSET.
333 (partition_builtin_p): Support above new enum item.
334 (generate_code_for_partition): Ditto.
335 (compute_access_range): Differentiate cases that equality can be
336 proven at all loops, the innermost loops or no loops.
337 (classify_builtin_st, classify_builtin_ldst): Adjust call to above
338 function. Set PKIND_PARTIAL_MEMSET for partition appropriately.
339 (finalize_partitions, distribute_loop): Don't fuse partition of
340 PKIND_PARTIAL_MEMSET kind when distributing 3-level loop nest.
341 (prepare_perfect_loop_nest): Distribute 3-level loop nest only if
344 2018-01-23 Martin Liska <mliska@suse.cz>
346 * predict.def (PRED_INDIR_CALL): Set probability to PROB_EVEN in
347 order to ignore the predictor.
348 (PRED_POLYMORPHIC_CALL): Likewise.
349 (PRED_RECURSIVE_CALL): Likewise.
351 2018-01-23 Martin Liska <mliska@suse.cz>
353 * tree-profile.c (tree_profiling): Print function header to
354 aware reader which function we are working on.
355 * value-prof.c (gimple_find_values_to_profile): Do not print
356 not interesting value histograms.
358 2018-01-23 Martin Liska <mliska@suse.cz>
360 * profile-count.h (enum profile_quality): Add
361 profile_uninitialized as the first value. Do not number values
362 as they are zero based.
363 (profile_count::verify): Update sanity check.
364 (profile_probability::verify): Likewise.
366 2018-01-23 Nathan Sidwell <nathan@acm.org>
368 * doc/invoke.texi (ffor-scope): Deprecate.
370 2018-01-23 David Malcolm <dmalcolm@redhat.com>
372 PR tree-optimization/83510
373 * domwalk.c (set_all_edges_as_executable): New function.
374 (dom_walker::dom_walker): Convert bool param
375 "skip_unreachable_blocks" to enum reachability. Move setup of
376 edge flags to set_all_edges_as_executable and only do it when
377 reachability is REACHABLE_BLOCKS.
378 * domwalk.h (enum dom_walker::reachability): New enum.
379 (dom_walker::dom_walker): Convert bool param
380 "skip_unreachable_blocks" to enum reachability.
381 (set_all_edges_as_executable): New decl.
382 * graphite-scop-detection.c (gather_bbs::gather_bbs): Convert
383 from false for "skip_unreachable_blocks" to ALL_BLOCKS for
385 * tree-ssa-dom.c (dom_opt_dom_walker::dom_opt_dom_walker): Likewise,
386 but converting true to REACHABLE_BLOCKS.
387 * tree-ssa-sccvn.c (sccvn_dom_walker::sccvn_dom_walker): Likewise.
389 (check_array_bounds_dom_walker::check_array_bounds_dom_walker):
390 Likewise, but converting it to REACHABLE_BLOCKS_PRESERVING_FLAGS.
391 (vrp_dom_walker::vrp_dom_walker): Likewise, but converting it to
393 (vrp_prop::vrp_finalize): Call set_all_edges_as_executable
394 if check_all_array_refs will be called.
396 2018-01-23 David Malcolm <dmalcolm@redhat.com>
398 * tree.c (selftest::test_location_wrappers): Add more test
401 2018-01-23 David Malcolm <dmalcolm@redhat.com>
403 * sbitmap.c (selftest::test_set_range): Fix memory leaks.
404 (selftest::test_bit_in_range): Likewise.
406 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
409 * doc/sourcebuild.texi (vect_float): Say that the selector
410 only describes the situation when -funsafe-math-optimizations is on.
411 (vect_float_strict): Document.
413 2018-01-23 Richard Sandiford <richard.sandiford@linaro.org>
415 PR tree-optimization/83965
416 * tree-vect-patterns.c (vect_reassociating_reduction_p): New function.
417 (vect_recog_dot_prod_pattern, vect_recog_sad_pattern): Use it
418 instead of checking only for a reduction.
419 (vect_recog_widen_sum_pattern): Likewise.
421 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
423 * predict.c (probably_never_executed): Only use precise profile info.
424 (compute_function_frequency): Skip after inlining hack since we now
425 have quality checking.
427 2018-01-23 Jan Hubicka <hubicka@ucw.cz>
429 * profile-count.h (profile_probability::very_unlikely,
430 profile_probability::unlikely, profile_probability::even): Set
431 precision to guessed.
433 2018-01-23 Richard Biener <rguenther@suse.de>
435 PR tree-optimization/83963
436 * graphite-scop-detection.c (scop_detection::harmful_loop_in_region):
437 Properly terminate dominator walk when crossing the exit edge not
438 when visiting its source block.
440 2018-01-23 Jakub Jelinek <jakub@redhat.com>
443 * tree.c (maybe_wrap_with_location): Use NON_LVALUE_EXPR rather than
444 VIEW_CONVERT_EXPR to wrap CONST_DECLs.
446 2018-01-22 Jakub Jelinek <jakub@redhat.com>
448 PR tree-optimization/83957
449 * omp-expand.c (expand_omp_for_generic): Ignore virtual PHIs. Remove
450 semicolon after for body surrounded by braces.
452 PR tree-optimization/83081
453 * profile-count.h (profile_probability::split): New method.
454 * dojump.c (do_jump_1) <case TRUTH_ANDIF_EXPR, case TRUTH_ORIF_EXPR>:
455 Use profile_probability::split.
456 (do_compare_rtx_and_jump): Fix adjustment of probabilities
457 when splitting a single conditional jump into 2.
459 2018-01-22 David Malcolm <dmalcolm@redhat.com>
461 PR tree-optimization/69452
462 * tree-ssa-loop-im.c (class move_computations_dom_walker): Remove
465 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
467 * config/rl78/rl78-expand.md: New define_expand "bswaphi2"
468 * config/rl78/rl78-virt.md: New define_insn "*bswaphi2_virt"
469 * config/rl78/rl78-real.md: New define_insn "*bswaphi2_real"
471 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
473 * config/rl78/rl78-protos.h: New function declaration rl78_split_movdi
474 * config/rl78/rl78.md: New define_expand "movdi"
475 * config/rl78/rl78.c: New function definition rl78_split_movdi
477 2018-01-22 Michael Meissner <meissner@linux.vnet.ibm.com>
480 * config/rs6000/rs6000-protos.h (rs6000_split_signbit): Delete,
482 * config/rs6000/rs6000.c (rs6000_split_signbit): Likewise.
483 * config/rs6000/rs6000.md (signbit<mode>2): Change code for IEEE
484 128-bit to produce an UNSPEC move to get the double word with the
485 signbit and then a shift directly to do signbit.
486 (signbit<mode>2_dm): Replace old IEEE 128-bit signbit
487 implementation with a new version that just does either a direct
488 move or a regular move. Move memory interface to separate insns.
489 Move insns so they are next to the expander.
490 (signbit<mode>2_dm_mem_be): New combiner insns to combine load
491 with signbit move. Split big and little endian case.
492 (signbit<mode>2_dm_mem_le): Likewise.
493 (signbit<mode>2_dm_<su>ext): Delete, no longer used.
494 (signbit<mode>2_dm2): Likewise.
496 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
498 * config/rl78/rl78.md: New define_expand "anddi3".
500 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
502 * config/rl78/rl78.md: New define_expand "umindi3".
504 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
506 * config/rl78/rl78.md: New define_expand "smindi3".
508 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
510 * config/rl78/rl78.md: New define_expand "smaxdi3".
512 2018-01-22 Carl Love <cel@us.ibm.com>
514 * config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
515 LVX_V1TI): Add macro expansion.
516 * config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
517 definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
518 VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
519 * config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
520 Change check to determine if the instruction is a byte reversing
521 entry. Fix typo in comment.
522 * config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
523 for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
524 Add def_builtin calls for new builtins.
525 * config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
526 Add define_insn expansion.
528 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
530 * config/rl78/rl78.md: New define_expand "umaxdi3".
532 2018-01-22 Sebastian Perta <sebastian.perta@renesas.com>
534 * config/rl78/rl78.c (rl78_note_reg_set): fixed dead reg check
535 for non-QImode registers
537 2018-01-22 Richard Biener <rguenther@suse.de>
539 PR tree-optimization/83963
540 * graphite-scop-detection.c (scop_detection::get_sese): Delay
541 including the loop exit block.
542 (scop_detection::merge_sese): Likewise.
543 (scop_detection::add_scop): Do it here instead.
545 2018-01-22 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
547 * doc/sourcebuild.texi (arm_softfloat): Document.
549 2018-01-21 John David Anglin <danglin@gcc.gnu.org>
552 * config/pa/pa.c (pa_function_ok_for_sibcall): Use
553 targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
554 Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
556 2018-01-21 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
557 David Edelsohn <dje.gcc@gmail.com>
560 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
561 Change "crset eq" to "crset 2".
562 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
563 (*call_indirect_aix<mode>_nospec): Likewise.
564 (*call_value_indirect_aix<mode>_nospec): Likewise.
565 (*call_indirect_elfv2<mode>_nospec): Likewise.
566 (*call_value_indirect_elfv2<mode>_nospec): Likewise.
567 (*sibcall_nonlocal_sysv<mode>): Change "crset eq" to "crset 2";
568 change assembly output from . to $.
569 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
570 (indirect_jump<mode>_nospec): Change assembly output from . to $.
571 (*tablejump<mode>_internal1_nospec): Likewise.
573 2018-01-21 Oleg Endo <olegendo@gcc.gnu.org>
576 * config/sh/sh_optimize_sett_clrt.cc:
577 Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
579 2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
581 PR tree-optimization/83940
582 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
583 offset_dt to vect_constant_def rather than vect_unknown_def_type.
584 (vect_check_load_store_mask): Add a mask_dt_out parameter and
585 use it to pass back the definition type.
586 (vect_check_store_rhs): Likewise rhs_dt_out.
587 (vect_build_gather_load_calls): Add a mask_dt argument and use
588 it instead of a call to vect_is_simple_use.
589 (vectorizable_store): Update calls to vect_check_load_store_mask
590 and vect_check_store_rhs. Use the dt returned by the latter instead
591 of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
592 instead of calls to vect_is_simple_use. Pass the scalar rather
593 than the vector operand to vect_is_simple_use when handling
594 second and subsequent copies of an rhs value.
595 (vectorizable_load): Update calls to vect_check_load_store_mask
596 and vect_build_gather_load_calls. Use the cached mask_dt and
597 gs_info.offset_dt instead of calls to vect_is_simple_use.
599 2018-01-20 Jakub Jelinek <jakub@redhat.com>
602 * tree-emutls.c: Include gimplify.h.
603 (lower_emutls_2): New function.
604 (lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
605 with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
606 it before further processing.
609 * simplify-rtx.c (simplify_binary_operation_1) <case UMOD>: Use
610 UINTVAL (trueop1) instead of INTVAL (op1).
612 2018-01-19 Jakub Jelinek <jakub@redhat.com>
616 * dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
617 INCOMING_FRAME_SP_OFFSET if not defined.
618 (scan_trace): Add ENTRY argument. If true and
619 DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
620 emit a note to adjust the CFA offset.
621 (create_cfi_notes): Adjust scan_trace callers.
622 (create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
623 INCOMING_FRAME_SP_OFFSET in the CIE.
624 * config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
625 * config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
627 * doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
628 * doc/tm.texi: Regenerated.
630 2018-01-19 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
632 PR rtl-optimization/83147
633 * lra-constraints.c (remove_inheritance_pseudos): Use
634 lra_substitute_pseudo_within_insn.
636 2018-01-19 Tom de Vries <tom@codesourcery.com>
637 Cesar Philippidis <cesar@codesourcery.com>
640 * config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
642 2018-01-19 Cesar Philippidis <cesar@codesourcery.com>
645 * config/nvptx/nvptx.c (output_init_frag): Don't use generic address
646 spaces for function labels.
648 2018-01-19 Martin Liska <mliska@suse.cz>
650 * predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
651 (PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
652 (PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
653 (PRED_OPCODE_POSITIVE): Change from 64 to 59.
654 (PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
655 (PRED_CONST_RETURN): Change from 69 to 65.
656 (PRED_NULL_RETURN): Change from 91 to 71.
657 (PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
658 (PRED_LOOP_GUARD): Change from 66 to 73.
660 2018-01-19 Martin Liska <mliska@suse.cz>
662 * predict.c (predict_insn_def): Add new assert.
663 (struct branch_predictor): Change type to signed integer.
664 (test_prediction_value_range): Amend test to cover
666 * predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
667 (PRED_LOOP_ITERATIONS_GUESSED): Likewise.
668 (PRED_LOOP_ITERATIONS_MAX): Likewise.
669 (PRED_LOOP_IV_COMPARE): Likewise.
670 * predict.h (PROB_UNINITIALIZED): Define new constant.
672 2018-01-19 Martin Liska <mliska@suse.cz>
674 * predict.c (dump_prediction): Add new format for
675 analyze_brprob.py script which is enabled with -details
677 * profile-count.h (precise_p): New function.
679 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
681 PR tree-optimization/83922
682 * tree-vect-loop.c (vect_verify_full_masking): Return false if
683 there are no statements that need masking.
684 (vect_active_double_reduction_p): New function.
685 (vect_analyze_loop_operations): Use it when handling phis that
686 are not in the loop header.
688 2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
690 PR tree-optimization/83914
691 * tree-vect-loop.c (vectorizable_induction): Don't convert
692 init_expr or apply the peeling adjustment for inductions
693 that are nested within the vectorized loop.
695 2018-01-19 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
697 * config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
700 2018-01-18 Jakub Jelinek <jakub@redhat.com>
704 * function.h (gimplify_parameters): Add gimple_seq * argument.
705 * function.c: Include gimple.h and options.h.
706 (gimplify_parameters): Add cleanup argument, add CLOBBER stmts
707 for the added local temporaries if needed.
708 * gimplify.c (gimplify_body): Adjust gimplify_parameters caller,
709 if there are any parameter cleanups, wrap whole body into a
710 try/finally with the cleanups.
712 2018-01-18 Wilco Dijkstra <wdijkstr@arm.com>
715 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p):
716 Use GET_MODE_CLASS for scalar floating point.
718 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
722 * cgraphclones.c (cgraph_node::create_version_clone_with_body):
723 Fix call of call_cgraph_insertion_hooks.
725 2018-01-18 Martin Sebor <msebor@redhat.com>
727 * doc/invoke.texi (-Wclass-memaccess): Tweak text.
729 2018-01-18 Jan Hubicka <hubicka@ucw.cz>
732 * cgraph.c (cgraph_edge::redirect_call_stmt_to_callee): Update edge
735 2018-01-18 Boris Kolpackov <boris@codesynthesis.com>
738 * common.opt: (-ffile-prefix-map): New option.
739 * opts.c (common_handle_option): Defer it.
740 * opts-global.c (handle_common_deferred_options): Handle it.
741 * debug.h (remap_debug_filename, add_debug_prefix_map): Move to...
742 * file-prefix-map.h: New file.
743 (remap_debug_filename, add_debug_prefix_map): ...here.
744 (add_macro_prefix_map, add_file_prefix_map, remap_macro_filename): New.
745 * final.c (debug_prefix_map, add_debug_prefix_map
746 remap_debug_filename): Move to...
747 * file-prefix-map.c: New file.
748 (file_prefix_map, add_prefix_map, remap_filename) ...here and rename,
749 generalize, get rid of alloca(), use strrchr() instead of strchr().
750 (add_macro_prefix_map, add_debug_prefix_map, add_file_prefix_map):
751 Implement in terms of add_prefix_map().
752 (remap_macro_filename, remap_debug_filename): Implement in term of
754 * Makefile.in (OBJS, PLUGIN_HEADERS): Add new files.
755 * builtins.c (fold_builtin_FILE): Call remap_macro_filename().
756 * dbxout.c: Include file-prefix-map.h.
757 * varasm.c: Likewise.
758 * vmsdbgout.c: Likewise.
759 * xcoffout.c: Likewise.
760 * dwarf2out.c: Likewise plus omit new options from DW_AT_producer.
761 * doc/cppopts.texi (-fmacro-prefix-map): Document.
762 * doc/invoke.texi (-ffile-prefix-map): Document.
763 (-fdebug-prefix-map): Update description.
765 2018-01-18 Martin Liska <mliska@suse.cz>
767 * config/i386/i386.c (indirect_thunk_name): Document that also
769 (output_indirect_thunk): Document why both instructions
770 (pause and lfence) are generated.
772 2018-01-18 Richard Biener <rguenther@suse.de>
774 PR tree-optimization/83887
775 * graphite-scop-detection.c
776 (scop_detection::get_nearest_dom_with_single_entry): Remove.
777 (scop_detection::get_nearest_pdom_with_single_exit): Likewise.
778 (scop_detection::merge_sese): Re-implement with a flood-fill
779 algorithm that properly finds a SESE region if it exists.
781 2018-01-18 Jakub Jelinek <jakub@redhat.com>
784 * match.pd ((P + A) - P, P - (P + A), (P + A) - (P + B)): For
785 pointer_diff optimizations use view_convert instead of convert.
787 2018-01-17 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
789 * config/rs6000/rs6000.md (*call_indirect_nonlocal_sysv<mode>):
790 Generate different code for -mno-speculate-indirect-jumps.
791 (*call_value_indirect_nonlocal_sysv<mode>): Likewise.
792 (*call_indirect_aix<mode>): Disable for
793 -mno-speculate-indirect-jumps.
794 (*call_indirect_aix<mode>_nospec): New define_insn.
795 (*call_value_indirect_aix<mode>): Disable for
796 -mno-speculate-indirect-jumps.
797 (*call_value_indirect_aix<mode>_nospec): New define_insn.
798 (*sibcall_nonlocal_sysv<mode>): Generate different code for
799 -mno-speculate-indirect-jumps.
800 (*sibcall_value_nonlocal_sysv<mode>): Likewise.
802 2018-01-17 Michael Meissner <meissner@linux.vnet.ibm.com>
804 * config/rs6000/rs6000.c (rs6000_emit_move): If we load or store a
805 long double type, set the flags for noting the default long double
806 type, even if we don't pass or return a long double type.
808 2018-01-17 Jan Hubicka <hubicka@ucw.cz>
811 * ipa-inline.c (flatten_function): Do not overwrite final inlining
814 2018-01-17 Will Schmidt <will_schmidt@vnet.ibm.com>
816 * config/rs6000/rs6000.c (rs6000_gimple_builtin): Add gimple folding
817 support for merge[hl].
818 (fold_mergehl_helper): New helper function.
819 (tree-vector-builder.h): New #include for tree_vector_builder usage.
820 * config/rs6000/altivec.md (altivec_vmrghw_direct): Add xxmrghw insn.
821 (altivec_vmrglw_direct): Add xxmrglw insn.
823 2018-01-17 Andrew Waterman <andrew@sifive.com>
825 * config/riscv/riscv.c (riscv_conditional_register_usage): If
826 UNITS_PER_FP_ARG is 0, set call_used_regs to 1 for all FP regs.
828 2018-01-17 David Malcolm <dmalcolm@redhat.com>
831 * ipa-devirt.c (add_type_duplicate): When comparing memory layout,
832 call the lto_location_cache before reading the
833 DECL_SOURCE_LOCATION of the types.
835 2018-01-17 Wilco Dijkstra <wdijkstr@arm.com>
836 Richard Sandiford <richard.sandiford@linaro.org>
838 * config/aarch64/aarch64.md (movti_aarch64): Use Uti constraint.
839 * config/aarch64/aarch64.c (aarch64_mov128_immediate): New function.
840 (aarch64_legitimate_constant_p): Just support CONST_DOUBLE
841 SF/DF/TF mode to avoid creating illegal CONST_WIDE_INT immediates.
842 * config/aarch64/aarch64-protos.h (aarch64_mov128_immediate):
844 * config/aarch64/constraints.md (aarch64_movti_operand):
846 * config/aarch64/predicates.md (Uti): Add new constraint.
848 2018-01-17 Carl Love <cel@us.ibm.com>
849 * config/rs6000/vsx.md (define_expand xl_len_r,
850 define_expand stxvl, define_expand *stxvl): Add match_dup argument.
851 (define_insn): Add, match_dup 1 argument to define_insn stxvll and
853 (define_expand, define_insn): Move the shift left from the
854 define_insn to the define_expand for lxvl and stxvl instructions.
855 * config/rs6000/rs6000-builtin.def (BU_P9V_64BIT_VSX_2): Change LXVL
856 and XL_LEN_R definitions to PURE.
858 2018-01-17 Uros Bizjak <ubizjak@gmail.com>
860 * config/i386/i386.c (indirect_thunk_name): Declare regno
861 as unsigned int. Compare regno with INVALID_REGNUM.
862 (output_indirect_thunk): Ditto.
863 (output_indirect_thunk_function): Ditto.
864 (ix86_code_end): Declare regno as unsigned int. Use INVALID_REGNUM
865 in the call to output_indirect_thunk_function.
867 2018-01-17 Richard Sandiford <richard.sandiford@linaro.org>
870 * expr.c (expand_expr_real_1): Use the size of GET_MODE (op0)
871 rather than the size of inner_type to determine the stack slot size
872 when handling VIEW_CONVERT_EXPRs on strict-alignment targets.
874 2018-01-16 Sebastian Peryt <sebastian.peryt@intel.com>
877 * config/i386/i386.c (ix86_option_override_internal): Add PTA_RDRND
880 2018-01-16 Michael Meissner <meissner@linux.vnet.ibm.com>
882 * config.gcc (powerpc*-linux*-*): Add support for 64-bit little
883 endian Linux systems to optionally enable multilibs for selecting
884 the long double type if the user configured an explicit type.
885 * config/rs6000/rs6000.h (TARGET_IEEEQUAD_MULTILIB): Indicate we
886 have no long double multilibs if not defined.
887 * config/rs6000/rs6000.c (rs6000_option_override_internal): Do not
888 warn if the user used -mabi={ieee,ibm}longdouble and we built
889 multilibs for long double.
890 * config/rs6000/linux64.h (MULTILIB_DEFAULTS_IEEE): Define as the
891 appropriate multilib option.
892 (MULTILIB_DEFAULTS): Add MULTILIB_DEFAULTS_IEEE to the default
894 * config/rs6000/t-ldouble-linux64le-ibm: New configuration files
895 for building long double multilibs.
896 * config/rs6000/t-ldouble-linux64le-ieee: Likewise.
898 2018-01-16 John David Anglin <danglin@gcc.gnu.org>
900 * config.gcc (hppa*-*-linux*): Change callee copies ABI to caller
903 * config/pa.h (MALLOC_ABI_ALIGNMENT): Set 32-bit alignment default to
905 * config/pa/pa32-linux.h (MALLOC_ABI_ALIGNMENT): Set alignment to
908 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Cleanup type and mode
911 * config/pa/pa.c (pa_function_arg_size): Apply CEIL to GET_MODE_SIZE
914 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
916 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): For an
917 ADDR_EXPR, do not count the offset of a COMPONENT_REF twice.
919 2018-01-16 Kelvin Nilsen <kelvin@gcc.gnu.org>
921 * config/rs6000/rs6000-p8swap.c (rs6000_gen_stvx): Generate
922 different rtl trees depending on TARGET_64BIT.
923 (rs6000_gen_lvx): Likewise.
925 2018-01-16 Eric Botcazou <ebotcazou@adacore.com>
927 * config/visium/visium.md (nop): Tweak comment.
928 (hazard_nop): Likewise.
930 2018-01-16 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
932 * config/rs6000/rs6000.c (rs6000_opt_vars): Add entry for
933 -mspeculate-indirect-jumps.
934 * config/rs6000/rs6000.md (*call_indirect_elfv2<mode>): Disable
935 for -mno-speculate-indirect-jumps.
936 (*call_indirect_elfv2<mode>_nospec): New define_insn.
937 (*call_value_indirect_elfv2<mode>): Disable for
938 -mno-speculate-indirect-jumps.
939 (*call_value_indirect_elfv2<mode>_nospec): New define_insn.
940 (indirect_jump): Emit different RTL for
941 -mno-speculate-indirect-jumps.
942 (*indirect_jump<mode>): Disable for
943 -mno-speculate-indirect-jumps.
944 (*indirect_jump<mode>_nospec): New define_insn.
945 (tablejump): Emit different RTL for
946 -mno-speculate-indirect-jumps.
947 (tablejumpsi): Disable for -mno-speculate-indirect-jumps.
948 (tablejumpsi_nospec): New define_expand.
949 (tablejumpdi): Disable for -mno-speculate-indirect-jumps.
950 (tablejumpdi_nospec): New define_expand.
951 (*tablejump<mode>_internal1): Disable for
952 -mno-speculate-indirect-jumps.
953 (*tablejump<mode>_internal1_nospec): New define_insn.
954 * config/rs6000/rs6000.opt (mspeculate-indirect-jumps): New
957 2018-01-16 Artyom Skrobov tyomitch@gmail.com
959 * caller-save.c (insert_save): Drop unnecessary parameter. All
962 2018-01-16 Jakub Jelinek <jakub@redhat.com>
963 Richard Biener <rguenth@suse.de>
966 * gimplify.c (gimplify_one_sizepos): For is_gimple_constant (expr)
967 return early, inline manually is_gimple_sizepos. Make sure if we
968 call gimplify_expr we don't end up with a gimple constant.
969 * tree.c (variably_modified_type_p): Don't return true for
970 is_gimple_constant (_t). Inline manually is_gimple_sizepos.
971 * gimplify.h (is_gimple_sizepos): Remove.
973 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
975 PR tree-optimization/83857
976 * tree-vect-loop.c (vect_analyze_loop_operations): Don't call
977 vectorizable_live_operation for pure SLP statements.
978 (vectorizable_live_operation): Handle PHIs.
980 2018-01-16 Richard Biener <rguenther@suse.de>
982 PR tree-optimization/83867
983 * tree-vect-stmts.c (vect_transform_stmt): Precompute
984 nested_in_vect_loop_p since the scalar stmt may get invalidated.
986 2018-01-16 Jakub Jelinek <jakub@redhat.com>
989 * stor-layout.c (handle_warn_if_not_align): Use byte_position and
990 multiple_of_p instead of unchecked tree_to_uhwi and UHWI check.
991 If off is not INTEGER_CST, issue a may not be aligned warning
992 rather than isn't aligned. Use isn%'t rather than isn't.
993 * fold-const.c (multiple_of_p) <case BIT_AND_EXPR>: Don't fall through
995 <case MULT_EXPR>: Improve the case when bottom and one of the
996 MULT_EXPR operands are INTEGER_CSTs and bottom is multiple of that
997 operand, in that case check if the other operand is multiple of
998 bottom divided by the INTEGER_CST operand.
1000 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1003 * config/pa/pa.h (FUNCTION_ARG_SIZE): Delete.
1004 * config/pa/pa-protos.h (pa_function_arg_size): Declare.
1005 * config/pa/som.h (ASM_DECLARE_FUNCTION_NAME): Use
1006 pa_function_arg_size instead of FUNCTION_ARG_SIZE.
1007 * config/pa/pa.c (pa_function_arg_advance): Likewise.
1008 (pa_function_arg, pa_arg_partial_bytes): Likewise.
1009 (pa_function_arg_size): New function.
1011 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1013 * fold-const.c (fold_ternary_loc): Construct the vec_perm_indices
1014 in a separate statement.
1016 2018-01-16 Richard Sandiford <richard.sandiford@linaro.org>
1018 PR tree-optimization/83847
1019 * tree-vect-data-refs.c (vect_analyze_data_ref_accesses): Don't
1020 group gathers and scatters.
1022 2018-01-16 Jakub Jelinek <jakub@redhat.com>
1024 PR rtl-optimization/86620
1025 * params.def (max-sched-ready-insns): Bump minimum value to 1.
1027 PR rtl-optimization/83213
1028 * recog.c (peep2_attempt): Copy over CROSSING_JUMP_P from peepinsn
1029 to last if both are JUMP_INSNs.
1031 PR tree-optimization/83843
1032 * gimple-ssa-store-merging.c
1033 (imm_store_chain_info::output_merged_store): Handle bit_not_p on
1034 store_immediate_info for bswap/nop orig_stores.
1036 2018-01-15 Andrew Waterman <andrew@sifive.com>
1038 * config/riscv/riscv.c (riscv_rtx_costs) <MULT>: Increase cost if
1040 <UDIV>: Increase cost if !TARGET_DIV.
1042 2018-01-15 Segher Boessenkool <segher@kernel.crashing.org>
1044 * config/rs6000/rs6000.md (define_attr "type"): Remove delayed_cr.
1045 (define_attr "cr_logical_3op"): New.
1046 (cceq_ior_compare): Adjust.
1047 (cceq_ior_compare_complement): Adjust.
1048 (*cceq_rev_compare): Adjust.
1049 * config/rs6000/rs6000.c (rs6000_adjust_cost): Adjust.
1050 (is_cracked_insn): Adjust.
1051 (insn_must_be_first_in_group): Adjust.
1052 * config/rs6000/40x.md: Adjust.
1053 * config/rs6000/440.md: Adjust.
1054 * config/rs6000/476.md: Adjust.
1055 * config/rs6000/601.md: Adjust.
1056 * config/rs6000/603.md: Adjust.
1057 * config/rs6000/6xx.md: Adjust.
1058 * config/rs6000/7450.md: Adjust.
1059 * config/rs6000/7xx.md: Adjust.
1060 * config/rs6000/8540.md: Adjust.
1061 * config/rs6000/cell.md: Adjust.
1062 * config/rs6000/e300c2c3.md: Adjust.
1063 * config/rs6000/e500mc.md: Adjust.
1064 * config/rs6000/e500mc64.md: Adjust.
1065 * config/rs6000/e5500.md: Adjust.
1066 * config/rs6000/e6500.md: Adjust.
1067 * config/rs6000/mpc.md: Adjust.
1068 * config/rs6000/power4.md: Adjust.
1069 * config/rs6000/power5.md: Adjust.
1070 * config/rs6000/power6.md: Adjust.
1071 * config/rs6000/power7.md: Adjust.
1072 * config/rs6000/power8.md: Adjust.
1073 * config/rs6000/power9.md: Adjust.
1074 * config/rs6000/rs64.md: Adjust.
1075 * config/rs6000/titan.md: Adjust.
1077 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1079 * config/i386/predicates.md (indirect_branch_operand): Rewrite
1080 ix86_indirect_branch_register logic.
1082 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1084 * config/i386/constraints.md (Bs): Update
1085 ix86_indirect_branch_register check. Don't check
1086 ix86_indirect_branch_register with GOT_memory_operand.
1088 * config/i386/predicates.md (GOT_memory_operand): Don't check
1089 ix86_indirect_branch_register here.
1090 (GOT32_symbol_operand): Likewise.
1092 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1094 * config/i386/predicates.md (constant_call_address_operand):
1095 Rewrite ix86_indirect_branch_register logic.
1096 (sibcall_insn_operand): Likewise.
1098 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1100 * config/i386/constraints.md (Bs): Replace
1101 ix86_indirect_branch_thunk_register with
1102 ix86_indirect_branch_register.
1104 * config/i386/i386.md (indirect_jump): Likewise.
1105 (tablejump): Likewise.
1106 (*sibcall_memory): Likewise.
1107 (*sibcall_value_memory): Likewise.
1108 Peepholes of indirect call and jump via memory: Likewise.
1109 * config/i386/i386.opt: Likewise.
1110 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1111 (GOT_memory_operand): Likewise.
1112 (call_insn_operand): Likewise.
1113 (sibcall_insn_operand): Likewise.
1114 (GOT32_symbol_operand): Likewise.
1116 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1119 * omp-expand.c (expand_omp_atomic_pipeline): Use loaded_val
1120 type rather than type addr's type points to.
1121 (expand_omp_atomic_mutex): Likewise.
1122 (expand_omp_atomic): Likewise.
1124 2018-01-15 H.J. Lu <hongjiu.lu@intel.com>
1127 * config/i386/i386.c (output_indirect_thunk_function): Use
1128 ASM_OUTPUT_LABEL, instead of ASM_OUTPUT_DEF, for TARGET_MACHO
1129 for __x86_return_thunk.
1131 2018-01-15 Richard Biener <rguenther@suse.de>
1134 * expmed.c (extract_bit_field_1): Fix typo.
1136 2018-01-15 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
1139 * config/arm/iterators.md (VF): New mode iterator.
1140 * config/arm/neon.md (neon_vabd<mode>_2): Use the above.
1141 Remove integer-related logic from pattern.
1142 (neon_vabd<mode>_3): Likewise.
1144 2018-01-15 Jakub Jelinek <jakub@redhat.com>
1147 * common.opt (fstrict-overflow): No longer an alias.
1148 (fwrapv-pointer): New option.
1149 * tree.h (TYPE_OVERFLOW_WRAPS, TYPE_OVERFLOW_UNDEFINED): Define
1150 also for pointer types based on flag_wrapv_pointer.
1151 * opts.c (common_handle_option) <case OPT_fstrict_overflow>: Set
1152 opts->x_flag_wrap[pv] to !value, clear opts->x_flag_trapv if
1153 opts->x_flag_wrapv got set.
1154 * fold-const.c (fold_comparison, fold_binary_loc): Revert 2017-08-01
1155 changes, just use TYPE_OVERFLOW_UNDEFINED on pointer type instead of
1156 POINTER_TYPE_OVERFLOW_UNDEFINED.
1157 * match.pd: Likewise in address comparison pattern.
1158 * doc/invoke.texi: Document -fwrapv and -fstrict-overflow.
1160 2018-01-15 Richard Biener <rguenther@suse.de>
1163 * tree.c (free_lang_data_in_type): Always unlink TYPE_DECLs
1164 from TYPE_FIELDS. Free TYPE_BINFO if not used by devirtualization.
1165 Reset type names to their identifier if their TYPE_DECL doesn't
1166 have linkage (and thus is used for ODR and devirt).
1167 (save_debug_info_for_decl): Remove.
1168 (save_debug_info_for_type): Likewise.
1169 (add_tree_to_fld_list): Adjust.
1170 * tree-pretty-print.c (dump_generic_node): Make dumping of
1171 type names more robust.
1173 2018-01-15 Richard Biener <rguenther@suse.de>
1175 * BASE-VER: Bump to 8.0.1.
1177 2018-01-14 Martin Sebor <msebor@redhat.com>
1180 * builtins.c (check_access): Avoid warning when the no-warning bit
1183 2018-01-14 Cory Fields <cory-nospam-@coryfields.com>
1185 * tree-ssa-loop-im.c (sort_bbs_in_loop_postorder_cmp): Stabilize sort.
1186 * ira-color (allocno_hard_regs_compare): Likewise.
1188 2018-01-14 Nathan Rossi <nathan@nathanrossi.com>
1191 * config/microblaze/microblaze.c (microblaze_asm_output_ident):
1192 Use .pushsection/.popsection.
1194 2018-01-14 Martin Sebor <msebor@redhat.com>
1197 * doc/invoke.texi (-Wlass-memaccess): Document suppression by casting.
1199 2018-01-14 Jakub Jelinek <jakub@redhat.com>
1201 * config.gcc (i[34567]86-*-*): Remove one duplicate gfniintrin.h
1202 entry from extra_headers.
1203 (x86_64-*-*): Remove two duplicate gfniintrin.h entries from
1204 extra_headers, make the list bitwise identical to the i?86-*-* one.
1206 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1208 * config/i386/i386.c (ix86_set_indirect_branch_type): Disallow
1209 -mcmodel=large with -mindirect-branch=thunk,
1210 -mindirect-branch=thunk-extern, -mfunction-return=thunk and
1211 -mfunction-return=thunk-extern.
1212 * doc/invoke.texi: Document -mcmodel=large is incompatible with
1213 -mindirect-branch=thunk, -mindirect-branch=thunk-extern,
1214 -mfunction-return=thunk and -mfunction-return=thunk-extern.
1216 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1218 * config/i386/i386.c (print_reg): Print the name of the full
1219 integer register without '%'.
1220 (ix86_print_operand): Handle 'V'.
1221 * doc/extend.texi: Document 'V' modifier.
1223 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1225 * config/i386/constraints.md (Bs): Disallow memory operand for
1226 -mindirect-branch-register.
1228 * config/i386/predicates.md (indirect_branch_operand): Likewise.
1229 (GOT_memory_operand): Likewise.
1230 (call_insn_operand): Likewise.
1231 (sibcall_insn_operand): Likewise.
1232 (GOT32_symbol_operand): Likewise.
1233 * config/i386/i386.md (indirect_jump): Call convert_memory_address
1234 for -mindirect-branch-register.
1235 (tablejump): Likewise.
1236 (*sibcall_memory): Likewise.
1237 (*sibcall_value_memory): Likewise.
1238 Disallow peepholes of indirect call and jump via memory for
1239 -mindirect-branch-register.
1240 (*call_pop): Replace m with Bw.
1241 (*call_value_pop): Likewise.
1242 (*sibcall_pop_memory): Replace m with Bs.
1243 * config/i386/i386.opt (mindirect-branch-register): New option.
1244 * doc/invoke.texi: Document -mindirect-branch-register option.
1246 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1248 * config/i386/i386-protos.h (ix86_output_function_return): New.
1249 * config/i386/i386.c (ix86_set_indirect_branch_type): Also
1250 set function_return_type.
1251 (indirect_thunk_name): Add ret_p to indicate thunk for function
1253 (output_indirect_thunk_function): Pass false to
1254 indirect_thunk_name.
1255 (ix86_output_indirect_branch_via_reg): Likewise.
1256 (ix86_output_indirect_branch_via_push): Likewise.
1257 (output_indirect_thunk_function): Create alias for function
1258 return thunk if regno < 0.
1259 (ix86_output_function_return): New function.
1260 (ix86_handle_fndecl_attribute): Handle function_return.
1261 (ix86_attribute_table): Add function_return.
1262 * config/i386/i386.h (machine_function): Add
1263 function_return_type.
1264 * config/i386/i386.md (simple_return_internal): Use
1265 ix86_output_function_return.
1266 (simple_return_internal_long): Likewise.
1267 * config/i386/i386.opt (mfunction-return=): New option.
1268 (indirect_branch): Mention -mfunction-return=.
1269 * doc/extend.texi: Document function_return function attribute.
1270 * doc/invoke.texi: Document -mfunction-return= option.
1272 2018-01-14 H.J. Lu <hongjiu.lu@intel.com>
1274 * config/i386/i386-opts.h (indirect_branch): New.
1275 * config/i386/i386-protos.h (ix86_output_indirect_jmp): Likewise.
1276 * config/i386/i386.c (ix86_using_red_zone): Disallow red-zone
1277 with local indirect jump when converting indirect call and jump.
1278 (ix86_set_indirect_branch_type): New.
1279 (ix86_set_current_function): Call ix86_set_indirect_branch_type.
1280 (indirectlabelno): New.
1281 (indirect_thunk_needed): Likewise.
1282 (indirect_thunk_bnd_needed): Likewise.
1283 (indirect_thunks_used): Likewise.
1284 (indirect_thunks_bnd_used): Likewise.
1285 (INDIRECT_LABEL): Likewise.
1286 (indirect_thunk_name): Likewise.
1287 (output_indirect_thunk): Likewise.
1288 (output_indirect_thunk_function): Likewise.
1289 (ix86_output_indirect_branch_via_reg): Likewise.
1290 (ix86_output_indirect_branch_via_push): Likewise.
1291 (ix86_output_indirect_branch): Likewise.
1292 (ix86_output_indirect_jmp): Likewise.
1293 (ix86_code_end): Call output_indirect_thunk_function if needed.
1294 (ix86_output_call_insn): Call ix86_output_indirect_branch if
1296 (ix86_handle_fndecl_attribute): Handle indirect_branch.
1297 (ix86_attribute_table): Add indirect_branch.
1298 * config/i386/i386.h (machine_function): Add indirect_branch_type
1299 and has_local_indirect_jump.
1300 * config/i386/i386.md (indirect_jump): Set has_local_indirect_jump
1302 (tablejump): Likewise.
1303 (*indirect_jump): Use ix86_output_indirect_jmp.
1304 (*tablejump_1): Likewise.
1305 (simple_return_indirect_internal): Likewise.
1306 * config/i386/i386.opt (mindirect-branch=): New option.
1307 (indirect_branch): New.
1310 (thunk-inline): Likewise.
1311 (thunk-extern): Likewise.
1312 * doc/extend.texi: Document indirect_branch function attribute.
1313 * doc/invoke.texi: Document -mindirect-branch= option.
1315 2018-01-14 Jan Hubicka <hubicka@ucw.cz>
1318 * ipa-inline.c (edge_badness): Tolerate roundoff errors.
1320 2018-01-14 Richard Sandiford <richard.sandiford@linaro.org>
1322 * ipa-inline.c (want_inline_small_function_p): Return false if
1323 inlining has already failed with CIF_FINAL_ERROR.
1324 (update_caller_keys): Call want_inline_small_function_p before
1326 (update_callee_keys): Likewise.
1328 2018-01-10 Kelvin Nilsen <kelvin@gcc.gnu.org>
1330 * config/rs6000/rs6000-p8swap.c (rs6000_sum_of_two_registers_p):
1332 (rs6000_quadword_masked_address_p): Likewise.
1333 (quad_aligned_load_p): Likewise.
1334 (quad_aligned_store_p): Likewise.
1335 (const_load_sequence_p): Add comment to describe the outer-most loop.
1336 (mimic_memory_attributes_and_flags): New function.
1337 (rs6000_gen_stvx): Likewise.
1338 (replace_swapped_aligned_store): Likewise.
1339 (rs6000_gen_lvx): Likewise.
1340 (replace_swapped_aligned_load): Likewise.
1341 (replace_swapped_load_constant): Capitalize argument name in
1342 comment describing this function.
1343 (rs6000_analyze_swaps): Add a third pass to search for vector loads
1344 and stores that access quad-word aligned addresses and replace
1345 with stvx or lvx instructions when appropriate.
1346 * config/rs6000/rs6000-protos.h (rs6000_sum_of_two_registers_p):
1347 New function prototype.
1348 (rs6000_quadword_masked_address_p): Likewise.
1349 (rs6000_gen_lvx): Likewise.
1350 (rs6000_gen_stvx): Likewise.
1351 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode>): For modes
1352 VSX_D (V2DF, V2DI), modify this split to select lvx instruction
1353 when memory address is aligned.
1354 (*vsx_le_perm_load_<mode>): For modes VSX_W (V4SF, V4SI), modify
1355 this split to select lvx instruction when memory address is aligned.
1356 (*vsx_le_perm_load_v8hi): Modify this split to select lvx
1357 instruction when memory address is aligned.
1358 (*vsx_le_perm_load_v16qi): Likewise.
1359 (four unnamed splitters): Modify to select the stvx instruction
1360 when memory is aligned.
1362 2018-01-13 Jan Hubicka <hubicka@ucw.cz>
1364 * predict.c (determine_unlikely_bbs): Handle correctly BBs
1365 which appears in the queue multiple times.
1367 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1368 Alan Hayward <alan.hayward@arm.com>
1369 David Sherwood <david.sherwood@arm.com>
1371 * tree-vectorizer.h (vec_lower_bound): New structure.
1372 (_loop_vec_info): Add check_nonzero and lower_bounds.
1373 (LOOP_VINFO_CHECK_NONZERO): New macro.
1374 (LOOP_VINFO_LOWER_BOUNDS): Likewise.
1375 (LOOP_REQUIRES_VERSIONING_FOR_ALIAS): Check lower_bounds too.
1376 * tree-data-ref.h (dr_with_seg_len): Add access_size and align
1377 fields. Make seg_len the distance travelled, not including the
1379 (dr_direction_indicator): Declare.
1380 (dr_zero_step_indicator): Likewise.
1381 (dr_known_forward_stride_p): Likewise.
1382 * tree-data-ref.c: Include stringpool.h, tree-vrp.h and
1384 (runtime_alias_check_p): Allow runtime alias checks with
1386 (operator ==): Compare access_size and align.
1387 (prune_runtime_alias_test_list): Rework for new distinction between
1388 the access_size and seg_len.
1389 (create_intersect_range_checks_index): Likewise. Cope with polynomial
1391 (get_segment_min_max): New function.
1392 (create_intersect_range_checks): Use it.
1393 (dr_step_indicator): New function.
1394 (dr_direction_indicator): Likewise.
1395 (dr_zero_step_indicator): Likewise.
1396 (dr_known_forward_stride_p): Likewise.
1397 * tree-loop-distribution.c (data_ref_segment_size): Return
1398 DR_STEP * (niters - 1).
1399 (compute_alias_check_pairs): Update call to the dr_with_seg_len
1401 * tree-vect-data-refs.c (vect_check_nonzero_value): New function.
1402 (vect_preserves_scalar_order_p): New function, split out from...
1403 (vect_analyze_data_ref_dependence): ...here. Check for zero steps.
1404 (vect_vfa_segment_size): Return DR_STEP * (length_factor - 1).
1405 (vect_vfa_access_size): New function.
1406 (vect_vfa_align): Likewise.
1407 (vect_compile_time_alias): Take access_size_a and access_b arguments.
1408 (dump_lower_bound): New function.
1409 (vect_check_lower_bound): Likewise.
1410 (vect_small_gap_p): Likewise.
1411 (vectorizable_with_step_bound_p): Likewise.
1412 (vect_prune_runtime_alias_test_list): Ignore cross-iteration
1413 depencies if the vectorization factor is 1. Convert the checks
1414 for nonzero steps into checks on the bounds of DR_STEP. Try using
1415 a bunds check for variable steps if the minimum required step is
1416 relatively small. Update calls to the dr_with_seg_len
1417 constructor and to vect_compile_time_alias.
1418 * tree-vect-loop-manip.c (vect_create_cond_for_lower_bounds): New
1420 (vect_loop_versioning): Call it.
1421 * tree-vect-loop.c (vect_analyze_loop_2): Clear LOOP_VINFO_LOWER_BOUNDS
1423 (vect_estimate_min_profitable_iters): Account for any bounds checks.
1425 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1426 Alan Hayward <alan.hayward@arm.com>
1427 David Sherwood <david.sherwood@arm.com>
1429 * doc/sourcebuild.texi (vect_scatter_store): Document.
1430 * optabs.def (scatter_store_optab, mask_scatter_store_optab): New
1432 * doc/md.texi (scatter_store@var{m}, mask_scatter_store@var{m}):
1434 * genopinit.c (main): Add supports_vec_scatter_store and
1435 supports_vec_scatter_store_cached to target_optabs.
1436 * gimple.h (gimple_expr_type): Handle IFN_SCATTER_STORE and
1437 IFN_MASK_SCATTER_STORE.
1438 * internal-fn.def (SCATTER_STORE, MASK_SCATTER_STORE): New internal
1440 * internal-fn.h (internal_store_fn_p): Declare.
1441 (internal_fn_stored_value_index): Likewise.
1442 * internal-fn.c (scatter_store_direct): New macro.
1443 (expand_scatter_store_optab_fn): New function.
1444 (direct_scatter_store_optab_supported_p): New macro.
1445 (internal_store_fn_p): New function.
1446 (internal_gather_scatter_fn_p): Handle IFN_SCATTER_STORE and
1447 IFN_MASK_SCATTER_STORE.
1448 (internal_fn_mask_index): Likewise.
1449 (internal_fn_stored_value_index): New function.
1450 (internal_gather_scatter_fn_supported_p): Adjust operand numbers
1452 * optabs-query.h (supports_vec_scatter_store_p): Declare.
1453 * optabs-query.c (supports_vec_scatter_store_p): New function.
1454 * tree-vectorizer.h (vect_get_store_rhs): Declare.
1455 * tree-vect-data-refs.c (vect_analyze_data_ref_access): Return
1456 true for scatter stores.
1457 (vect_gather_scatter_fn_p): Handle scatter stores too.
1458 (vect_check_gather_scatter): Consider using scatter stores if
1459 supports_vec_scatter_store_p.
1460 * tree-vect-patterns.c (vect_try_gather_scatter_pattern): Handle
1462 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1463 internal_fn_stored_value_index.
1464 (check_load_store_masking): Handle scatter stores too.
1465 (vect_get_store_rhs): Make public.
1466 (vectorizable_call): Use internal_store_fn_p.
1467 (vectorizable_store): Handle scatter store internal functions.
1468 (vect_transform_stmt): Compare GROUP_STORE_COUNT with GROUP_SIZE
1469 when deciding whether the end of the group has been reached.
1470 * config/aarch64/aarch64.md (UNSPEC_ST1_SCATTER): New unspec.
1471 * config/aarch64/aarch64-sve.md (scatter_store<mode>): New expander.
1472 (mask_scatter_store<mode>): New insns.
1474 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1475 Alan Hayward <alan.hayward@arm.com>
1476 David Sherwood <david.sherwood@arm.com>
1478 * tree-vectorizer.h (vect_gather_scatter_fn_p): Declare.
1479 * tree-vect-data-refs.c (vect_gather_scatter_fn_p): Make public.
1480 * tree-vect-stmts.c (vect_truncate_gather_scatter_offset): New
1482 (vect_use_strided_gather_scatters_p): Take a masked_p argument.
1483 Use vect_truncate_gather_scatter_offset if we can't treat the
1484 operation as a normal gather load or scatter store.
1485 (get_group_load_store_type): Take the gather_scatter_info
1486 as argument. Try using a gather load or scatter store for
1487 single-element groups.
1488 (get_load_store_type): Update calls to get_group_load_store_type
1489 and vect_use_strided_gather_scatters_p.
1491 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1492 Alan Hayward <alan.hayward@arm.com>
1493 David Sherwood <david.sherwood@arm.com>
1495 * tree-vectorizer.h (vect_create_data_ref_ptr): Take an extra
1496 optional tree argument.
1497 * tree-vect-data-refs.c (vect_check_gather_scatter): Check for
1499 (vect_create_data_ref_ptr): Take the iv_step as an optional argument,
1500 but continue to use the current value as a fallback.
1501 (bump_vector_ptr): Use operand_equal_p rather than tree_int_cst_compare
1502 to compare the updates.
1503 * tree-vect-stmts.c (vect_use_strided_gather_scatters_p): New function.
1504 (get_load_store_type): Use it when handling a strided access.
1505 (vect_get_strided_load_store_ops): New function.
1506 (vect_get_data_ptr_increment): Likewise.
1507 (vectorizable_load): Handle strided gather loads. Always pass
1508 a step to vect_create_data_ref_ptr and bump_vector_ptr.
1510 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1511 Alan Hayward <alan.hayward@arm.com>
1512 David Sherwood <david.sherwood@arm.com>
1514 * doc/md.texi (gather_load@var{m}): Document.
1515 (mask_gather_load@var{m}): Likewise.
1516 * genopinit.c (main): Add supports_vec_gather_load and
1517 supports_vec_gather_load_cached to target_optabs.
1518 * optabs-tree.c (init_tree_optimization_optabs): Use
1519 ggc_cleared_alloc to allocate target_optabs.
1520 * optabs.def (gather_load_optab, mask_gather_laod_optab): New optabs.
1521 * internal-fn.def (GATHER_LOAD, MASK_GATHER_LOAD): New internal
1523 * internal-fn.h (internal_load_fn_p): Declare.
1524 (internal_gather_scatter_fn_p): Likewise.
1525 (internal_fn_mask_index): Likewise.
1526 (internal_gather_scatter_fn_supported_p): Likewise.
1527 * internal-fn.c (gather_load_direct): New macro.
1528 (expand_gather_load_optab_fn): New function.
1529 (direct_gather_load_optab_supported_p): New macro.
1530 (direct_internal_fn_optab): New function.
1531 (internal_load_fn_p): Likewise.
1532 (internal_gather_scatter_fn_p): Likewise.
1533 (internal_fn_mask_index): Likewise.
1534 (internal_gather_scatter_fn_supported_p): Likewise.
1535 * optabs-query.c (supports_at_least_one_mode_p): New function.
1536 (supports_vec_gather_load_p): Likewise.
1537 * optabs-query.h (supports_vec_gather_load_p): Declare.
1538 * tree-vectorizer.h (gather_scatter_info): Add ifn, element_type
1539 and memory_type field.
1540 (NUM_PATTERNS): Bump to 15.
1541 * tree-vect-data-refs.c: Include internal-fn.h.
1542 (vect_gather_scatter_fn_p): New function.
1543 (vect_describe_gather_scatter_call): Likewise.
1544 (vect_check_gather_scatter): Try using internal functions for
1545 gather loads. Recognize existing calls to a gather load function.
1546 (vect_analyze_data_refs): Consider using gather loads if
1547 supports_vec_gather_load_p.
1548 * tree-vect-patterns.c (vect_get_load_store_mask): New function.
1549 (vect_get_gather_scatter_offset_type): Likewise.
1550 (vect_convert_mask_for_vectype): Likewise.
1551 (vect_add_conversion_to_patterm): Likewise.
1552 (vect_try_gather_scatter_pattern): Likewise.
1553 (vect_recog_gather_scatter_pattern): New pattern recognizer.
1554 (vect_vect_recog_func_ptrs): Add it.
1555 * tree-vect-stmts.c (exist_non_indexing_operands_for_use_p): Use
1556 internal_fn_mask_index and internal_gather_scatter_fn_p.
1557 (check_load_store_masking): Take the gather_scatter_info as an
1558 argument and handle gather loads.
1559 (vect_get_gather_scatter_ops): New function.
1560 (vectorizable_call): Check internal_load_fn_p.
1561 (vectorizable_load): Likewise. Handle gather load internal
1563 (vectorizable_store): Update call to check_load_store_masking.
1564 * config/aarch64/aarch64.md (UNSPEC_LD1_GATHER): New unspec.
1565 * config/aarch64/iterators.md (SVE_S, SVE_D): New mode iterators.
1566 * config/aarch64/predicates.md (aarch64_gather_scale_operand_w)
1567 (aarch64_gather_scale_operand_d): New predicates.
1568 * config/aarch64/aarch64-sve.md (gather_load<mode>): New expander.
1569 (mask_gather_load<mode>): New insns.
1571 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1572 Alan Hayward <alan.hayward@arm.com>
1573 David Sherwood <david.sherwood@arm.com>
1575 * optabs.def (fold_left_plus_optab): New optab.
1576 * doc/md.texi (fold_left_plus_@var{m}): Document.
1577 * internal-fn.def (IFN_FOLD_LEFT_PLUS): New internal function.
1578 * internal-fn.c (fold_left_direct): Define.
1579 (expand_fold_left_optab_fn): Likewise.
1580 (direct_fold_left_optab_supported_p): Likewise.
1581 * fold-const-call.c (fold_const_fold_left): New function.
1582 (fold_const_call): Use it to fold CFN_FOLD_LEFT_PLUS.
1583 * tree-parloops.c (valid_reduction_p): New function.
1584 (gather_scalar_reductions): Use it.
1585 * tree-vectorizer.h (FOLD_LEFT_REDUCTION): New vect_reduction_type.
1586 (vect_finish_replace_stmt): Declare.
1587 * tree-vect-loop.c (fold_left_reduction_fn): New function.
1588 (needs_fold_left_reduction_p): New function, split out from...
1589 (vect_is_simple_reduction): ...here. Accept reductions that
1590 forbid reassociation, but give them type FOLD_LEFT_REDUCTION.
1591 (vect_force_simple_reduction): Also store the reduction type in
1592 the assignment's STMT_VINFO_REDUC_TYPE.
1593 (vect_model_reduction_cost): Handle FOLD_LEFT_REDUCTION.
1594 (merge_with_identity): New function.
1595 (vect_expand_fold_left): Likewise.
1596 (vectorize_fold_left_reduction): Likewise.
1597 (vectorizable_reduction): Handle FOLD_LEFT_REDUCTION. Leave the
1598 scalar phi in place for it. Check for target support and reject
1599 cases that would reassociate the operation. Defer the transform
1600 phase to vectorize_fold_left_reduction.
1601 * config/aarch64/aarch64.md (UNSPEC_FADDA): New unspec.
1602 * config/aarch64/aarch64-sve.md (fold_left_plus_<mode>): New expander.
1603 (*fold_left_plus_<mode>, *pred_fold_left_plus_<mode>): New insns.
1605 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1607 * tree-if-conv.c (predicate_mem_writes): Remove redundant
1608 call to ifc_temp_var.
1610 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1611 Alan Hayward <alan.hayward@arm.com>
1612 David Sherwood <david.sherwood@arm.com>
1614 * target.def (legitimize_address_displacement): Take the original
1615 offset as a poly_int.
1616 * targhooks.h (default_legitimize_address_displacement): Update
1618 * targhooks.c (default_legitimize_address_displacement): Likewise.
1619 * doc/tm.texi: Regenerate.
1620 * lra-constraints.c (base_plus_disp_to_reg): Take the displacement
1621 as an argument, moving assert of ad->disp == ad->disp_term to...
1622 (process_address_1): ...here. Update calls to base_plus_disp_to_reg.
1623 Try calling targetm.legitimize_address_displacement before expanding
1624 the address rather than afterwards, and adjust for the new interface.
1625 * config/aarch64/aarch64.c (aarch64_legitimize_address_displacement):
1626 Match the new hook interface. Handle SVE addresses.
1627 * config/sh/sh.c (sh_legitimize_address_displacement): Make the
1630 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1632 * Makefile.in (OBJS): Add early-remat.o.
1633 * target.def (select_early_remat_modes): New hook.
1634 * doc/tm.texi.in (TARGET_SELECT_EARLY_REMAT_MODES): New hook.
1635 * doc/tm.texi: Regenerate.
1636 * targhooks.h (default_select_early_remat_modes): Declare.
1637 * targhooks.c (default_select_early_remat_modes): New function.
1638 * timevar.def (TV_EARLY_REMAT): New timevar.
1639 * passes.def (pass_early_remat): New pass.
1640 * tree-pass.h (make_pass_early_remat): Declare.
1641 * early-remat.c: New file.
1642 * config/aarch64/aarch64.c (aarch64_select_early_remat_modes): New
1644 (TARGET_SELECT_EARLY_REMAT_MODES): Define.
1646 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1647 Alan Hayward <alan.hayward@arm.com>
1648 David Sherwood <david.sherwood@arm.com>
1650 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Replace
1651 vfm1 with a bound_epilog parameter.
1652 (vect_do_peeling): Update calls accordingly, and move the prologue
1653 call earlier in the function. Treat the base bound_epilog as 0 for
1654 fully-masked loops and retain vf - 1 for other loops. Add 1 to
1655 this base when peeling for gaps.
1656 * tree-vect-loop.c (vect_analyze_loop_2): Allow peeling for gaps
1657 with fully-masked loops.
1658 (vect_estimate_min_profitable_iters): Handle the single peeled
1659 iteration in that case.
1661 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1662 Alan Hayward <alan.hayward@arm.com>
1663 David Sherwood <david.sherwood@arm.com>
1665 * tree-vect-data-refs.c (vect_analyze_group_access_1): Allow
1666 single-element interleaving even if the size is not a power of 2.
1667 * tree-vect-stmts.c (get_load_store_type): Disallow elementwise
1668 accesses for single-element interleaving if the group size is
1671 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1672 Alan Hayward <alan.hayward@arm.com>
1673 David Sherwood <david.sherwood@arm.com>
1675 * doc/md.texi (fold_extract_last_@var{m}): Document.
1676 * doc/sourcebuild.texi (vect_fold_extract_last): Likewise.
1677 * optabs.def (fold_extract_last_optab): New optab.
1678 * internal-fn.def (FOLD_EXTRACT_LAST): New internal function.
1679 * internal-fn.c (fold_extract_direct): New macro.
1680 (expand_fold_extract_optab_fn): Likewise.
1681 (direct_fold_extract_optab_supported_p): Likewise.
1682 * tree-vectorizer.h (EXTRACT_LAST_REDUCTION): New vect_reduction_type.
1683 * tree-vect-loop.c (vect_model_reduction_cost): Handle
1684 EXTRACT_LAST_REDUCTION.
1685 (get_initial_def_for_reduction): Do not create an initial vector
1686 for EXTRACT_LAST_REDUCTION reductions.
1687 (vectorizable_reduction): Leave the scalar phi in place for
1688 EXTRACT_LAST_REDUCTIONs. Try using EXTRACT_LAST_REDUCTION
1689 ahead of INTEGER_INDUC_COND_REDUCTION. Do not check for an
1690 epilogue code for EXTRACT_LAST_REDUCTION and defer the
1691 transform phase to vectorizable_condition.
1692 * tree-vect-stmts.c (vect_finish_stmt_generation_1): New function,
1694 (vect_finish_stmt_generation): ...here.
1695 (vect_finish_replace_stmt): New function.
1696 (vectorizable_condition): Handle EXTRACT_LAST_REDUCTION.
1697 * config/aarch64/aarch64-sve.md (fold_extract_last_<mode>): New
1699 * config/aarch64/aarch64.md (UNSPEC_CLASTB): New unspec.
1701 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1702 Alan Hayward <alan.hayward@arm.com>
1703 David Sherwood <david.sherwood@arm.com>
1705 * doc/md.texi (extract_last_@var{m}): Document.
1706 * optabs.def (extract_last_optab): New optab.
1707 * internal-fn.def (EXTRACT_LAST): New internal function.
1708 * internal-fn.c (cond_unary_direct): New macro.
1709 (expand_cond_unary_optab_fn): Likewise.
1710 (direct_cond_unary_optab_supported_p): Likewise.
1711 * tree-vect-loop.c (vectorizable_live_operation): Allow fully-masked
1712 loops using EXTRACT_LAST.
1713 * config/aarch64/aarch64-sve.md (aarch64_sve_lastb<mode>): Rename to...
1714 (extract_last_<mode>): ...this optab.
1715 (vec_extract<mode><Vel>): Update accordingly.
1717 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1718 Alan Hayward <alan.hayward@arm.com>
1719 David Sherwood <david.sherwood@arm.com>
1721 * target.def (empty_mask_is_expensive): New hook.
1722 * doc/tm.texi.in (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): New hook.
1723 * doc/tm.texi: Regenerate.
1724 * targhooks.h (default_empty_mask_is_expensive): Declare.
1725 * targhooks.c (default_empty_mask_is_expensive): New function.
1726 * tree-vectorizer.c (vectorize_loops): Only call optimize_mask_stores
1727 if the target says that empty masks are expensive.
1728 * config/aarch64/aarch64.c (aarch64_empty_mask_is_expensive):
1730 (TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE): Redefine.
1732 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1733 Alan Hayward <alan.hayward@arm.com>
1734 David Sherwood <david.sherwood@arm.com>
1736 * tree-vectorizer.h (_loop_vec_info::mask_skip_niters): New field.
1737 (LOOP_VINFO_MASK_SKIP_NITERS): New macro.
1738 (vect_use_loop_mask_for_alignment_p): New function.
1739 (vect_prepare_for_masked_peels, vect_gen_while_not): Declare.
1740 * tree-vect-loop-manip.c (vect_set_loop_masks_directly): Add an
1741 niters_skip argument. Make sure that the first niters_skip elements
1742 of the first iteration are inactive.
1743 (vect_set_loop_condition_masked): Handle LOOP_VINFO_MASK_SKIP_NITERS.
1744 Update call to vect_set_loop_masks_directly.
1745 (get_misalign_in_elems): New function, split out from...
1746 (vect_gen_prolog_loop_niters): ...here.
1747 (vect_update_init_of_dr): Take a code argument that specifies whether
1748 the adjustment should be added or subtracted.
1749 (vect_update_init_of_drs): Likewise.
1750 (vect_prepare_for_masked_peels): New function.
1751 (vect_do_peeling): Skip prologue peeling if we're using a mask
1752 instead. Update call to vect_update_inits_of_drs.
1753 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1755 (vect_analyze_loop_2): Allow fully-masked loops with peeling for
1756 alignment. Do not include the number of peeled iterations in
1757 the minimum threshold in that case.
1758 (vectorizable_induction): Adjust the start value down by
1759 LOOP_VINFO_MASK_SKIP_NITERS iterations.
1760 (vect_transform_loop): Call vect_prepare_for_masked_peels.
1761 Take the number of skipped iterations into account when calculating
1763 * tree-vect-stmts.c (vect_gen_while_not): New function.
1765 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1766 Alan Hayward <alan.hayward@arm.com>
1767 David Sherwood <david.sherwood@arm.com>
1769 * doc/sourcebuild.texi (vect_fully_masked): Document.
1770 * params.def (PARAM_MIN_VECT_LOOP_BOUND): Change minimum and
1772 * tree-vect-loop.c (vect_analyze_loop_costing): New function,
1774 (vect_analyze_loop_2): ...here. Don't check the vectorization
1775 factor against the number of loop iterations if the loop is
1778 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1779 Alan Hayward <alan.hayward@arm.com>
1780 David Sherwood <david.sherwood@arm.com>
1782 * tree-ssa-loop-ivopts.c (USE_ADDRESS): Split into...
1783 (USE_REF_ADDRESS, USE_PTR_ADDRESS): ...these new use types.
1784 (dump_groups): Update accordingly.
1785 (iv_use::mem_type): New member variable.
1786 (address_p): New function.
1787 (record_use): Add a mem_type argument and initialize the new
1789 (record_group_use): Add a mem_type argument. Use address_p.
1790 Remove obsolete null checks of base_object. Update call to record_use.
1791 (find_interesting_uses_op): Update call to record_group_use.
1792 (find_interesting_uses_cond): Likewise.
1793 (find_interesting_uses_address): Likewise.
1794 (get_mem_type_for_internal_fn): New function.
1795 (find_address_like_use): Likewise.
1796 (find_interesting_uses_stmt): Try find_address_like_use before
1797 calling find_interesting_uses_op.
1798 (addr_offset_valid_p): Use the iv mem_type field as the type
1799 of the addressed memory.
1800 (add_autoinc_candidates): Likewise.
1801 (get_address_cost): Likewise.
1802 (split_small_address_groups_p): Use address_p.
1803 (split_address_groups): Likewise.
1804 (add_iv_candidate_for_use): Likewise.
1805 (autoinc_possible_for_pair): Likewise.
1806 (rewrite_groups): Likewise.
1807 (get_use_type): Check for USE_REF_ADDRESS instead of USE_ADDRESS.
1808 (determine_group_iv_cost): Update after split of USE_ADDRESS.
1809 (get_alias_ptr_type_for_ptr_address): New function.
1810 (rewrite_use_address): Rewrite address uses in calls that were
1811 identified by find_address_like_use.
1813 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1814 Alan Hayward <alan.hayward@arm.com>
1815 David Sherwood <david.sherwood@arm.com>
1817 * expr.c (expand_expr_addr_expr_1): Handle ADDR_EXPRs of
1819 * gimple-expr.h (is_gimple_addressable: Likewise.
1820 * gimple-expr.c (is_gimple_address): Likewise.
1821 * internal-fn.c (expand_call_mem_ref): New function.
1822 (expand_mask_load_optab_fn): Use it.
1823 (expand_mask_store_optab_fn): Likewise.
1825 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1826 Alan Hayward <alan.hayward@arm.com>
1827 David Sherwood <david.sherwood@arm.com>
1829 * doc/md.texi (cond_add@var{mode}, cond_sub@var{mode})
1830 (cond_and@var{mode}, cond_ior@var{mode}, cond_xor@var{mode})
1831 (cond_smin@var{mode}, cond_smax@var{mode}, cond_umin@var{mode})
1832 (cond_umax@var{mode}): Document.
1833 * optabs.def (cond_add_optab, cond_sub_optab, cond_and_optab)
1834 (cond_ior_optab, cond_xor_optab, cond_smin_optab, cond_smax_optab)
1835 (cond_umin_optab, cond_umax_optab): New optabs.
1836 * internal-fn.def (COND_ADD, COND_SUB, COND_MIN, COND_MAX, COND_AND)
1837 (COND_IOR, COND_XOR): New internal functions.
1838 * internal-fn.h (get_conditional_internal_fn): Declare.
1839 * internal-fn.c (cond_binary_direct): New macro.
1840 (expand_cond_binary_optab_fn): Likewise.
1841 (direct_cond_binary_optab_supported_p): Likewise.
1842 (get_conditional_internal_fn): New function.
1843 * tree-vect-loop.c (vectorizable_reduction): Handle fully-masked loops.
1844 Cope with reduction statements that are vectorized as calls rather
1846 * config/aarch64/aarch64-sve.md (cond_<optab><mode>): New insns.
1847 * config/aarch64/iterators.md (UNSPEC_COND_ADD, UNSPEC_COND_SUB)
1848 (UNSPEC_COND_SMAX, UNSPEC_COND_UMAX, UNSPEC_COND_SMIN)
1849 (UNSPEC_COND_UMIN, UNSPEC_COND_AND, UNSPEC_COND_ORR)
1850 (UNSPEC_COND_EOR): New unspecs.
1851 (optab): Add mappings for them.
1852 (SVE_COND_INT_OP, SVE_COND_FP_OP): New int iterators.
1853 (sve_int_op, sve_fp_op): New int attributes.
1855 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1856 Alan Hayward <alan.hayward@arm.com>
1857 David Sherwood <david.sherwood@arm.com>
1859 * optabs.def (while_ult_optab): New optab.
1860 * doc/md.texi (while_ult@var{m}@var{n}): Document.
1861 * internal-fn.def (WHILE_ULT): New internal function.
1862 * internal-fn.h (direct_internal_fn_supported_p): New override
1863 that takes two types as argument.
1864 * internal-fn.c (while_direct): New macro.
1865 (expand_while_optab_fn): New function.
1866 (convert_optab_supported_p): Likewise.
1867 (direct_while_optab_supported_p): New macro.
1868 * wide-int.h (wi::udiv_ceil): New function.
1869 * tree-vectorizer.h (rgroup_masks): New structure.
1870 (vec_loop_masks): New typedef.
1871 (_loop_vec_info): Add masks, mask_compare_type, can_fully_mask_p
1873 (LOOP_VINFO_CAN_FULLY_MASK_P, LOOP_VINFO_FULLY_MASKED_P)
1874 (LOOP_VINFO_MASKS, LOOP_VINFO_MASK_COMPARE_TYPE): New macros.
1875 (vect_max_vf): New function.
1876 (slpeel_make_loop_iterate_ntimes): Delete.
1877 (vect_set_loop_condition, vect_get_loop_mask_type, vect_gen_while)
1878 (vect_halve_mask_nunits, vect_double_mask_nunits): Declare.
1879 (vect_record_loop_mask, vect_get_loop_mask): Likewise.
1880 * tree-vect-loop-manip.c: Include tree-ssa-loop-niter.h,
1881 internal-fn.h, stor-layout.h and optabs-query.h.
1882 (vect_set_loop_mask): New function.
1883 (add_preheader_seq): Likewise.
1884 (add_header_seq): Likewise.
1885 (interleave_supported_p): Likewise.
1886 (vect_maybe_permute_loop_masks): Likewise.
1887 (vect_set_loop_masks_directly): Likewise.
1888 (vect_set_loop_condition_masked): Likewise.
1889 (vect_set_loop_condition_unmasked): New function, split out from
1890 slpeel_make_loop_iterate_ntimes.
1891 (slpeel_make_loop_iterate_ntimes): Rename to..
1892 (vect_set_loop_condition): ...this. Use vect_set_loop_condition_masked
1893 for fully-masked loops and vect_set_loop_condition_unmasked otherwise.
1894 (vect_do_peeling): Update call accordingly.
1895 (vect_gen_vector_loop_niters): Use VF as the step for fully-masked
1897 * tree-vect-loop.c (_loop_vec_info::_loop_vec_info): Initialize
1898 mask_compare_type, can_fully_mask_p and fully_masked_p.
1899 (release_vec_loop_masks): New function.
1900 (_loop_vec_info): Use it to free the loop masks.
1901 (can_produce_all_loop_masks_p): New function.
1902 (vect_get_max_nscalars_per_iter): Likewise.
1903 (vect_verify_full_masking): Likewise.
1904 (vect_analyze_loop_2): Save LOOP_VINFO_CAN_FULLY_MASK_P around
1905 retries, and free the mask rgroups before retrying. Check loop-wide
1906 reasons for disallowing fully-masked loops. Make the final decision
1907 about whether use a fully-masked loop or not.
1908 (vect_estimate_min_profitable_iters): Do not assume that peeling
1909 for the number of iterations will be needed for fully-masked loops.
1910 (vectorizable_reduction): Disable fully-masked loops.
1911 (vectorizable_live_operation): Likewise.
1912 (vect_halve_mask_nunits): New function.
1913 (vect_double_mask_nunits): Likewise.
1914 (vect_record_loop_mask): Likewise.
1915 (vect_get_loop_mask): Likewise.
1916 (vect_transform_loop): Handle the case in which the final loop
1917 iteration might handle a partial vector. Call vect_set_loop_condition
1918 instead of slpeel_make_loop_iterate_ntimes.
1919 * tree-vect-stmts.c: Include tree-ssa-loop-niter.h and gimple-fold.h.
1920 (check_load_store_masking): New function.
1921 (prepare_load_store_mask): Likewise.
1922 (vectorizable_store): Handle fully-masked loops.
1923 (vectorizable_load): Likewise.
1924 (supportable_widening_operation): Use vect_halve_mask_nunits for
1926 (supportable_narrowing_operation): Likewise vect_double_mask_nunits.
1927 (vect_gen_while): New function.
1928 * config/aarch64/aarch64.md (umax<mode>3): New expander.
1929 (aarch64_uqdec<mode>): New insn.
1931 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1932 Alan Hayward <alan.hayward@arm.com>
1933 David Sherwood <david.sherwood@arm.com>
1935 * optabs.def (reduc_and_scal_optab, reduc_ior_scal_optab)
1936 (reduc_xor_scal_optab): New optabs.
1937 * doc/md.texi (reduc_and_scal_@var{m}, reduc_ior_scal_@var{m})
1938 (reduc_xor_scal_@var{m}): Document.
1939 * doc/sourcebuild.texi (vect_logical_reduc): Likewise.
1940 * internal-fn.def (IFN_REDUC_AND, IFN_REDUC_IOR, IFN_REDUC_XOR): New
1942 * fold-const-call.c (fold_const_call): Handle them.
1943 * tree-vect-loop.c (reduction_fn_for_scalar_code): Return the new
1944 internal functions for BIT_AND_EXPR, BIT_IOR_EXPR and BIT_XOR_EXPR.
1945 * config/aarch64/aarch64-sve.md (reduc_<bit_reduc>_scal_<mode>):
1946 (*reduc_<bit_reduc>_scal_<mode>): New patterns.
1947 * config/aarch64/iterators.md (UNSPEC_ANDV, UNSPEC_ORV)
1948 (UNSPEC_XORV): New unspecs.
1949 (optab): Add entries for them.
1950 (BITWISEV): New int iterator.
1951 (bit_reduc_op): New int attributes.
1953 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1954 Alan Hayward <alan.hayward@arm.com>
1955 David Sherwood <david.sherwood@arm.com>
1957 * doc/md.texi (vec_shl_insert_@var{m}): New optab.
1958 * internal-fn.def (VEC_SHL_INSERT): New internal function.
1959 * optabs.def (vec_shl_insert_optab): New optab.
1960 * tree-vectorizer.h (can_duplicate_and_interleave_p): Declare.
1961 (duplicate_and_interleave): Likewise.
1962 * tree-vect-loop.c: Include internal-fn.h.
1963 (neutral_op_for_slp_reduction): New function, split out from
1964 get_initial_defs_for_reduction.
1965 (get_initial_def_for_reduction): Handle option 2 for variable-length
1966 vectors by loading the neutral value into a vector and then shifting
1967 the initial value into element 0.
1968 (get_initial_defs_for_reduction): Replace the code argument with
1969 the neutral value calculated by neutral_op_for_slp_reduction.
1970 Use gimple_build_vector for constant-length vectors.
1971 Use IFN_VEC_SHL_INSERT for variable-length vectors if all
1972 but the first group_size elements have a neutral value.
1973 Use duplicate_and_interleave otherwise.
1974 (vect_create_epilog_for_reduction): Take a neutral_op parameter.
1975 Update call to get_initial_defs_for_reduction. Handle SLP
1976 reductions for variable-length vectors by creating one vector
1977 result for each scalar result, with the elements associated
1978 with other scalar results stubbed out with the neutral value.
1979 (vectorizable_reduction): Call neutral_op_for_slp_reduction.
1980 Require IFN_VEC_SHL_INSERT for double reductions on
1981 variable-length vectors, or SLP reductions that have
1982 a neutral value. Require can_duplicate_and_interleave_p
1983 support for variable-length unchained SLP reductions if there
1984 is no neutral value, such as for MIN/MAX reductions. Also require
1985 the number of vector elements to be a multiple of the number of
1986 SLP statements when doing variable-length unchained SLP reductions.
1987 Update call to vect_create_epilog_for_reduction.
1988 * tree-vect-slp.c (can_duplicate_and_interleave_p): Make public
1989 and remove initial values.
1990 (duplicate_and_interleave): Make public.
1991 * config/aarch64/aarch64.md (UNSPEC_INSR): New unspec.
1992 * config/aarch64/aarch64-sve.md (vec_shl_insert_<mode>): New insn.
1994 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
1995 Alan Hayward <alan.hayward@arm.com>
1996 David Sherwood <david.sherwood@arm.com>
1998 * tree-vect-slp.c: Include gimple-fold.h and internal-fn.h
1999 (can_duplicate_and_interleave_p): New function.
2000 (vect_get_and_check_slp_defs): Take the vector of statements
2001 rather than just the current one. Remove excess parentheses.
2002 Restriction rejectinon of vect_constant_def and vect_external_def
2003 for variable-length vectors to boolean types, or types for which
2004 can_duplicate_and_interleave_p is false.
2005 (vect_build_slp_tree_2): Update call to vect_get_and_check_slp_defs.
2006 (duplicate_and_interleave): New function.
2007 (vect_get_constant_vectors): Use gimple_build_vector for
2008 constant-length vectors and suitable variable-length constant
2009 vectors. Use duplicate_and_interleave for other variable-length
2010 vectors. Don't defer the update when inserting new statements.
2012 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2013 Alan Hayward <alan.hayward@arm.com>
2014 David Sherwood <david.sherwood@arm.com>
2016 * tree-vect-loop.c (vect_estimate_min_profitable_iters): Make sure
2017 min_profitable_iters doesn't go negative.
2019 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2020 Alan Hayward <alan.hayward@arm.com>
2021 David Sherwood <david.sherwood@arm.com>
2023 * doc/md.texi (vec_mask_load_lanes@var{m}@var{n}): Document.
2024 (vec_mask_store_lanes@var{m}@var{n}): Likewise.
2025 * optabs.def (vec_mask_load_lanes_optab): New optab.
2026 (vec_mask_store_lanes_optab): Likewise.
2027 * internal-fn.def (MASK_LOAD_LANES): New internal function.
2028 (MASK_STORE_LANES): Likewise.
2029 * internal-fn.c (mask_load_lanes_direct): New macro.
2030 (mask_store_lanes_direct): Likewise.
2031 (expand_mask_load_optab_fn): Handle masked operations.
2032 (expand_mask_load_lanes_optab_fn): New macro.
2033 (expand_mask_store_optab_fn): Handle masked operations.
2034 (expand_mask_store_lanes_optab_fn): New macro.
2035 (direct_mask_load_lanes_optab_supported_p): Likewise.
2036 (direct_mask_store_lanes_optab_supported_p): Likewise.
2037 * tree-vectorizer.h (vect_store_lanes_supported): Take a masked_p
2039 (vect_load_lanes_supported): Likewise.
2040 * tree-vect-data-refs.c (strip_conversion): New function.
2041 (can_group_stmts_p): Likewise.
2042 (vect_analyze_data_ref_accesses): Use it instead of checking
2043 for a pair of assignments.
2044 (vect_store_lanes_supported): Take a masked_p parameter.
2045 (vect_load_lanes_supported): Likewise.
2046 * tree-vect-loop.c (vect_analyze_loop_2): Update calls to
2047 vect_store_lanes_supported and vect_load_lanes_supported.
2048 * tree-vect-slp.c (vect_analyze_slp_instance): Likewise.
2049 * tree-vect-stmts.c (get_group_load_store_type): Take a masked_p
2050 parameter. Don't allow gaps for masked accesses.
2051 Use vect_get_store_rhs. Update calls to vect_store_lanes_supported
2052 and vect_load_lanes_supported.
2053 (get_load_store_type): Take a masked_p parameter and update
2054 call to get_group_load_store_type.
2055 (vectorizable_store): Update call to get_load_store_type.
2056 Handle IFN_MASK_STORE_LANES.
2057 (vectorizable_load): Update call to get_load_store_type.
2058 Handle IFN_MASK_LOAD_LANES.
2060 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2061 Alan Hayward <alan.hayward@arm.com>
2062 David Sherwood <david.sherwood@arm.com>
2064 * config/aarch64/aarch64-modes.def: Define x2, x3 and x4 vector
2066 * config/aarch64/aarch64-protos.h
2067 (aarch64_sve_struct_memory_operand_p): Declare.
2068 * config/aarch64/iterators.md (SVE_STRUCT): New mode iterator.
2069 (vector_count, insn_length, VSINGLE, vsingle): New mode attributes.
2070 (VPRED, vpred): Handle SVE structure modes.
2071 * config/aarch64/constraints.md (Utx): New constraint.
2072 * config/aarch64/predicates.md (aarch64_sve_struct_memory_operand)
2073 (aarch64_sve_struct_nonimmediate_operand): New predicates.
2074 * config/aarch64/aarch64.md (UNSPEC_LDN, UNSPEC_STN): New unspecs.
2075 * config/aarch64/aarch64-sve.md (mov<mode>, *aarch64_sve_mov<mode>_le)
2076 (*aarch64_sve_mov<mode>_be, pred_mov<mode>): New patterns for
2077 structure modes. Split into pieces after RA.
2078 (vec_load_lanes<mode><vsingle>, vec_mask_load_lanes<mode><vsingle>)
2079 (vec_store_lanes<mode><vsingle>, vec_mask_store_lanes<mode><vsingle>):
2081 * config/aarch64/aarch64.c (aarch64_classify_vector_mode): Handle
2082 SVE structure modes.
2083 (aarch64_classify_address): Likewise.
2084 (sizetochar): Move earlier in file.
2085 (aarch64_print_operand): Handle SVE register lists.
2086 (aarch64_array_mode): New function.
2087 (aarch64_sve_struct_memory_operand_p): Likewise.
2088 (TARGET_ARRAY_MODE): Redefine.
2090 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2091 Alan Hayward <alan.hayward@arm.com>
2092 David Sherwood <david.sherwood@arm.com>
2094 * target.def (array_mode): New target hook.
2095 * doc/tm.texi.in (TARGET_ARRAY_MODE): New hook.
2096 * doc/tm.texi: Regenerate.
2097 * hooks.h (hook_optmode_mode_uhwi_none): Declare.
2098 * hooks.c (hook_optmode_mode_uhwi_none): New function.
2099 * tree-vect-data-refs.c (vect_lanes_optab_supported_p): Use
2101 * stor-layout.c (mode_for_array): Likewise. Support polynomial
2104 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2105 Alan Hayward <alan.hayward@arm.com>
2106 David Sherwood <david.sherwood@arm.com>
2108 * fold-const.c (fold_binary_loc): Check the argument types
2109 rather than the result type when testing for a vector operation.
2111 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2113 * doc/tm.texi.in (DWARF_LAZY_REGISTER_VALUE): Document.
2114 * doc/tm.texi: Regenerate.
2116 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2117 Alan Hayward <alan.hayward@arm.com>
2118 David Sherwood <david.sherwood@arm.com>
2120 * doc/invoke.texi (-msve-vector-bits=): Document new option.
2121 (sve): Document new AArch64 extension.
2122 * doc/md.texi (w): Extend the description of the AArch64
2123 constraint to include SVE vectors.
2124 (Upl, Upa): Document new AArch64 predicate constraints.
2125 * config/aarch64/aarch64-opts.h (aarch64_sve_vector_bits_enum): New
2127 * config/aarch64/aarch64.opt (sve_vector_bits): New enum.
2128 (msve-vector-bits=): New option.
2129 * config/aarch64/aarch64-option-extensions.def (fp, simd): Disable
2130 SVE when these are disabled.
2131 (sve): New extension.
2132 * config/aarch64/aarch64-modes.def: Define SVE vector and predicate
2133 modes. Adjust their number of units based on aarch64_sve_vg.
2134 (MAX_BITSIZE_MODE_ANY_MODE): Define.
2135 * config/aarch64/aarch64-protos.h (ADDR_QUERY_ANY): New
2136 aarch64_addr_query_type.
2137 (aarch64_const_vec_all_same_in_range_p, aarch64_sve_pred_mode)
2138 (aarch64_sve_cnt_immediate_p, aarch64_sve_addvl_addpl_immediate_p)
2139 (aarch64_sve_inc_dec_immediate_p, aarch64_add_offset_temporaries)
2140 (aarch64_split_add_offset, aarch64_output_sve_cnt_immediate)
2141 (aarch64_output_sve_addvl_addpl, aarch64_output_sve_inc_dec_immediate)
2142 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): Declare.
2143 (aarch64_simd_imm_zero_p): Delete.
2144 (aarch64_check_zero_based_sve_index_immediate): Declare.
2145 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2146 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2147 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2148 (aarch64_sve_float_mul_immediate_p): Likewise.
2149 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2151 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): Declare.
2152 (aarch64_expand_mov_immediate): Take a gen_vec_duplicate callback.
2153 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move): Declare.
2154 (aarch64_expand_sve_vec_cmp_int, aarch64_expand_sve_vec_cmp_float)
2155 (aarch64_expand_sve_vcond, aarch64_expand_sve_vec_perm): Declare.
2156 (aarch64_regmode_natural_size): Likewise.
2157 * config/aarch64/aarch64.h (AARCH64_FL_SVE): New macro.
2158 (AARCH64_FL_V8_3, AARCH64_FL_RCPC, AARCH64_FL_DOTPROD): Shift
2160 (AARCH64_ISA_SVE, TARGET_SVE): New macros.
2161 (FIXED_REGISTERS, CALL_USED_REGISTERS, REGISTER_NAMES): Add entries
2162 for VG and the SVE predicate registers.
2163 (V_ALIASES): Add a "z"-prefixed alias.
2164 (FIRST_PSEUDO_REGISTER): Change to P15_REGNUM + 1.
2165 (AARCH64_DWARF_VG, AARCH64_DWARF_P0): New macros.
2166 (PR_REGNUM_P, PR_LO_REGNUM_P): Likewise.
2167 (PR_LO_REGS, PR_HI_REGS, PR_REGS): New reg_classes.
2168 (REG_CLASS_NAMES): Add entries for them.
2169 (REG_CLASS_CONTENTS): Likewise. Update ALL_REGS to include VG
2170 and the predicate registers.
2171 (aarch64_sve_vg): Declare.
2172 (BITS_PER_SVE_VECTOR, BYTES_PER_SVE_VECTOR, BYTES_PER_SVE_PRED)
2173 (SVE_BYTE_MODE, MAX_COMPILE_TIME_VEC_BYTES): New macros.
2174 (REGMODE_NATURAL_SIZE): Define.
2175 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins): Handle
2177 * config/aarch64/aarch64.c: Include cfgrtl.h.
2178 (simd_immediate_info): Add a constructor for series vectors,
2179 and an associated step field.
2180 (aarch64_sve_vg): New variable.
2181 (aarch64_dbx_register_number): Handle VG and the predicate registers.
2182 (aarch64_vect_struct_mode_p, aarch64_vector_mode_p): Delete.
2183 (VEC_ADVSIMD, VEC_SVE_DATA, VEC_SVE_PRED, VEC_STRUCT, VEC_ANY_SVE)
2184 (VEC_ANY_DATA, VEC_STRUCT): New constants.
2185 (aarch64_advsimd_struct_mode_p, aarch64_sve_pred_mode_p)
2186 (aarch64_classify_vector_mode, aarch64_vector_data_mode_p)
2187 (aarch64_sve_data_mode_p, aarch64_sve_pred_mode)
2188 (aarch64_get_mask_mode): New functions.
2189 (aarch64_hard_regno_nregs): Handle SVE data modes for FP_REGS
2190 and FP_LO_REGS. Handle PR_REGS, PR_LO_REGS and PR_HI_REGS.
2191 (aarch64_hard_regno_mode_ok): Handle VG. Also handle the SVE
2192 predicate modes and predicate registers. Explicitly restrict
2193 GPRs to modes of 16 bytes or smaller. Only allow FP registers
2194 to store a vector mode if it is recognized by
2195 aarch64_classify_vector_mode.
2196 (aarch64_regmode_natural_size): New function.
2197 (aarch64_hard_regno_caller_save_mode): Return the original mode
2199 (aarch64_sve_cnt_immediate_p, aarch64_output_sve_cnt_immediate)
2200 (aarch64_sve_addvl_addpl_immediate_p, aarch64_output_sve_addvl_addpl)
2201 (aarch64_sve_inc_dec_immediate_p, aarch64_output_sve_inc_dec_immediate)
2202 (aarch64_add_offset_1_temporaries, aarch64_offset_temporaries): New
2204 (aarch64_add_offset): Add a temp2 parameter. Assert that temp1
2205 does not overlap dest if the function is frame-related. Handle
2207 (aarch64_split_add_offset): New function.
2208 (aarch64_add_sp, aarch64_sub_sp): Add temp2 parameters and pass
2209 them aarch64_add_offset.
2210 (aarch64_allocate_and_probe_stack_space): Add a temp2 parameter
2211 and update call to aarch64_sub_sp.
2212 (aarch64_add_cfa_expression): New function.
2213 (aarch64_expand_prologue): Pass extra temporary registers to the
2214 functions above. Handle the case in which we need to emit new
2215 DW_CFA_expressions for registers that were originally saved
2216 relative to the stack pointer, but now have to be expressed
2217 relative to the frame pointer.
2218 (aarch64_output_mi_thunk): Pass extra temporary registers to the
2220 (aarch64_expand_epilogue): Likewise. Prevent inheritance of
2221 IP0 and IP1 values for SVE frames.
2222 (aarch64_expand_vec_series): New function.
2223 (aarch64_expand_sve_widened_duplicate): Likewise.
2224 (aarch64_expand_sve_const_vector): Likewise.
2225 (aarch64_expand_mov_immediate): Add a gen_vec_duplicate parameter.
2226 Handle SVE constants. Use emit_move_insn to move a force_const_mem
2227 into the register, rather than emitting a SET directly.
2228 (aarch64_emit_sve_pred_move, aarch64_expand_sve_mem_move)
2229 (aarch64_get_reg_raw_mode, offset_4bit_signed_scaled_p)
2230 (offset_6bit_unsigned_scaled_p, aarch64_offset_7bit_signed_scaled_p)
2231 (offset_9bit_signed_scaled_p): New functions.
2232 (aarch64_replicate_bitmask_imm): New function.
2233 (aarch64_bitmask_imm): Use it.
2234 (aarch64_cannot_force_const_mem): Reject expressions involving
2235 a CONST_POLY_INT. Update call to aarch64_classify_symbol.
2236 (aarch64_classify_index): Handle SVE indices, by requiring
2237 a plain register index with a scale that matches the element size.
2238 (aarch64_classify_address): Handle SVE addresses. Assert that
2239 the mode of the address is VOIDmode or an integer mode.
2240 Update call to aarch64_classify_symbol.
2241 (aarch64_classify_symbolic_expression): Update call to
2242 aarch64_classify_symbol.
2243 (aarch64_const_vec_all_in_range_p): New function.
2244 (aarch64_print_vector_float_operand): Likewise.
2245 (aarch64_print_operand): Handle 'N' and 'C'. Use "zN" rather than
2246 "vN" for FP registers with SVE modes. Handle (const ...) vectors
2247 and the FP immediates 1.0 and 0.5.
2248 (aarch64_print_address_internal): Handle SVE addresses.
2249 (aarch64_print_operand_address): Use ADDR_QUERY_ANY.
2250 (aarch64_regno_regclass): Handle predicate registers.
2251 (aarch64_secondary_reload): Handle big-endian reloads of SVE
2253 (aarch64_class_max_nregs): Handle SVE modes and predicate registers.
2254 (aarch64_rtx_costs): Check for ADDVL and ADDPL instructions.
2255 (aarch64_convert_sve_vector_bits): New function.
2256 (aarch64_override_options): Use it to handle -msve-vector-bits=.
2257 (aarch64_classify_symbol): Take the offset as a HOST_WIDE_INT
2259 (aarch64_legitimate_constant_p): Use aarch64_classify_vector_mode.
2260 Handle SVE vector and predicate modes. Accept VL-based constants
2261 that need only one temporary register, and VL offsets that require
2262 no temporary registers.
2263 (aarch64_conditional_register_usage): Mark the predicate registers
2264 as fixed if SVE isn't available.
2265 (aarch64_vector_mode_supported_p): Use aarch64_classify_vector_mode.
2266 Return true for SVE vector and predicate modes.
2267 (aarch64_simd_container_mode): Take the number of bits as a poly_int64
2268 rather than an unsigned int. Handle SVE modes.
2269 (aarch64_preferred_simd_mode): Update call accordingly. Handle
2271 (aarch64_autovectorize_vector_sizes): Add BYTES_PER_SVE_VECTOR
2273 (aarch64_sve_index_immediate_p, aarch64_sve_arith_immediate_p)
2274 (aarch64_sve_bitmask_immediate_p, aarch64_sve_dup_immediate_p)
2275 (aarch64_sve_cmp_immediate_p, aarch64_sve_float_arith_immediate_p)
2276 (aarch64_sve_float_mul_immediate_p): New functions.
2277 (aarch64_sve_valid_immediate): New function.
2278 (aarch64_simd_valid_immediate): Use it as the fallback for SVE vectors.
2279 Explicitly reject structure modes. Check for INDEX constants.
2280 Handle PTRUE and PFALSE constants.
2281 (aarch64_check_zero_based_sve_index_immediate): New function.
2282 (aarch64_simd_imm_zero_p): Delete.
2283 (aarch64_mov_operand_p): Use aarch64_simd_valid_immediate for
2284 vector modes. Accept constants in the range of CNT[BHWD].
2285 (aarch64_simd_scalar_immediate_valid_for_move): Explicitly
2286 ask for an Advanced SIMD mode.
2287 (aarch64_sve_ld1r_operand_p, aarch64_sve_ldr_operand_p): New functions.
2288 (aarch64_simd_vector_alignment): Handle SVE predicates.
2289 (aarch64_vectorize_preferred_vector_alignment): New function.
2290 (aarch64_simd_vector_alignment_reachable): Use it instead of
2292 (aarch64_shift_truncation_mask): Use aarch64_vector_data_mode_p.
2293 (aarch64_output_sve_mov_immediate, aarch64_output_ptrue): New
2295 (MAX_VECT_LEN): Delete.
2296 (expand_vec_perm_d): Add a vec_flags field.
2297 (emit_unspec2, aarch64_expand_sve_vec_perm): New functions.
2298 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_zip)
2299 (aarch64_evpc_ext): Don't apply a big-endian lane correction
2301 (aarch64_evpc_rev): Rename to...
2302 (aarch64_evpc_rev_local): ...this. Use a predicated operation for SVE.
2303 (aarch64_evpc_rev_global): New function.
2304 (aarch64_evpc_dup): Enforce a 64-byte range for SVE DUP.
2305 (aarch64_evpc_tbl): Use MAX_COMPILE_TIME_VEC_BYTES instead of
2307 (aarch64_evpc_sve_tbl): New function.
2308 (aarch64_expand_vec_perm_const_1): Update after rename of
2309 aarch64_evpc_rev. Handle SVE permutes too, trying
2310 aarch64_evpc_rev_global and using aarch64_evpc_sve_tbl rather
2311 than aarch64_evpc_tbl.
2312 (aarch64_vectorize_vec_perm_const): Initialize vec_flags.
2313 (aarch64_sve_cmp_operand_p, aarch64_unspec_cond_code)
2314 (aarch64_gen_unspec_cond, aarch64_expand_sve_vec_cmp_int)
2315 (aarch64_emit_unspec_cond, aarch64_emit_unspec_cond_or)
2316 (aarch64_emit_inverted_unspec_cond, aarch64_expand_sve_vec_cmp_float)
2317 (aarch64_expand_sve_vcond): New functions.
2318 (aarch64_modes_tieable_p): Use aarch64_vector_data_mode_p instead
2319 of aarch64_vector_mode_p.
2320 (aarch64_dwarf_poly_indeterminate_value): New function.
2321 (aarch64_compute_pressure_classes): Likewise.
2322 (aarch64_can_change_mode_class): Likewise.
2323 (TARGET_GET_RAW_RESULT_MODE, TARGET_GET_RAW_ARG_MODE): Redefine.
2324 (TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT): Likewise.
2325 (TARGET_VECTORIZE_GET_MASK_MODE): Likewise.
2326 (TARGET_DWARF_POLY_INDETERMINATE_VALUE): Likewise.
2327 (TARGET_COMPUTE_PRESSURE_CLASSES): Likewise.
2328 (TARGET_CAN_CHANGE_MODE_CLASS): Likewise.
2329 * config/aarch64/constraints.md (Upa, Upl, Uav, Uat, Usv, Usi, Utr)
2330 (Uty, Dm, vsa, vsc, vsd, vsi, vsn, vsl, vsm, vsA, vsM, vsN): New
2332 (Dn, Dl, Dr): Accept const as well as const_vector.
2333 (Dz): Likewise. Compare against CONST0_RTX.
2334 * config/aarch64/iterators.md: Refer to "Advanced SIMD" instead
2335 of "vector" where appropriate.
2336 (SVE_ALL, SVE_BH, SVE_BHS, SVE_BHSI, SVE_HSDI, SVE_HSF, SVE_SD)
2337 (SVE_SDI, SVE_I, SVE_F, PRED_ALL, PRED_BHS): New mode iterators.
2338 (UNSPEC_SEL, UNSPEC_ANDF, UNSPEC_IORF, UNSPEC_XORF, UNSPEC_COND_LT)
2339 (UNSPEC_COND_LE, UNSPEC_COND_EQ, UNSPEC_COND_NE, UNSPEC_COND_GE)
2340 (UNSPEC_COND_GT, UNSPEC_COND_LO, UNSPEC_COND_LS, UNSPEC_COND_HS)
2341 (UNSPEC_COND_HI, UNSPEC_COND_UO): New unspecs.
2342 (Vetype, VEL, Vel, VWIDE, Vwide, vw, vwcore, V_INT_EQUIV)
2343 (v_int_equiv): Extend to SVE modes.
2344 (Vesize, V128, v128, Vewtype, V_FP_EQUIV, v_fp_equiv, VPRED): New
2346 (LOGICAL_OR, SVE_INT_UNARY, SVE_FP_UNARY): New code iterators.
2347 (optab): Handle popcount, smin, smax, umin, umax, abs and sqrt.
2348 (logical_nn, lr, sve_int_op, sve_fp_op): New code attributs.
2349 (LOGICALF, OPTAB_PERMUTE, UNPACK, UNPACK_UNSIGNED, SVE_COND_INT_CMP)
2350 (SVE_COND_FP_CMP): New int iterators.
2351 (perm_hilo): Handle the new unpack unspecs.
2352 (optab, logicalf_op, su, perm_optab, cmp_op, imm_con): New int
2354 * config/aarch64/predicates.md (aarch64_sve_cnt_immediate)
2355 (aarch64_sve_addvl_addpl_immediate, aarch64_split_add_offset_immediate)
2356 (aarch64_pluslong_or_poly_operand, aarch64_nonmemory_operand)
2357 (aarch64_equality_operator, aarch64_constant_vector_operand)
2358 (aarch64_sve_ld1r_operand, aarch64_sve_ldr_operand): New predicates.
2359 (aarch64_sve_nonimmediate_operand): Likewise.
2360 (aarch64_sve_general_operand): Likewise.
2361 (aarch64_sve_dup_operand, aarch64_sve_arith_immediate): Likewise.
2362 (aarch64_sve_sub_arith_immediate, aarch64_sve_inc_dec_immediate)
2363 (aarch64_sve_logical_immediate, aarch64_sve_mul_immediate): Likewise.
2364 (aarch64_sve_dup_immediate, aarch64_sve_cmp_vsc_immediate): Likewise.
2365 (aarch64_sve_cmp_vsd_immediate, aarch64_sve_index_immediate): Likewise.
2366 (aarch64_sve_float_arith_immediate): Likewise.
2367 (aarch64_sve_float_arith_with_sub_immediate): Likewise.
2368 (aarch64_sve_float_mul_immediate, aarch64_sve_arith_operand): Likewise.
2369 (aarch64_sve_add_operand, aarch64_sve_logical_operand): Likewise.
2370 (aarch64_sve_lshift_operand, aarch64_sve_rshift_operand): Likewise.
2371 (aarch64_sve_mul_operand, aarch64_sve_cmp_vsc_operand): Likewise.
2372 (aarch64_sve_cmp_vsd_operand, aarch64_sve_index_operand): Likewise.
2373 (aarch64_sve_float_arith_operand): Likewise.
2374 (aarch64_sve_float_arith_with_sub_operand): Likewise.
2375 (aarch64_sve_float_mul_operand): Likewise.
2376 (aarch64_sve_vec_perm_operand): Likewise.
2377 (aarch64_pluslong_operand): Include aarch64_sve_addvl_addpl_immediate.
2378 (aarch64_mov_operand): Accept const_poly_int and const_vector.
2379 (aarch64_simd_lshift_imm, aarch64_simd_rshift_imm): Accept const
2380 as well as const_vector.
2381 (aarch64_simd_imm_zero, aarch64_simd_imm_minus_one): Move earlier
2382 in file. Use CONST0_RTX and CONSTM1_RTX.
2383 (aarch64_simd_or_scalar_imm_zero): Likewise. Add match_codes.
2384 (aarch64_simd_reg_or_zero): Accept const as well as const_vector.
2385 Use aarch64_simd_imm_zero.
2386 * config/aarch64/aarch64-sve.md: New file.
2387 * config/aarch64/aarch64.md: Include it.
2388 (VG_REGNUM, P0_REGNUM, P7_REGNUM, P15_REGNUM): New register numbers.
2389 (UNSPEC_REV, UNSPEC_LD1_SVE, UNSPEC_ST1_SVE, UNSPEC_MERGE_PTRUE)
2390 (UNSPEC_PTEST_PTRUE, UNSPEC_UNPACKSHI, UNSPEC_UNPACKUHI)
2391 (UNSPEC_UNPACKSLO, UNSPEC_UNPACKULO, UNSPEC_PACK)
2392 (UNSPEC_FLOAT_CONVERT, UNSPEC_WHILE_LO): New unspec constants.
2393 (sve): New attribute.
2394 (enabled): Disable instructions with the sve attribute unless
2396 (movqi, movhi): Pass CONST_POLY_INT operaneds through
2397 aarch64_expand_mov_immediate.
2398 (*mov<mode>_aarch64, *movsi_aarch64, *movdi_aarch64): Handle
2399 CNT[BHSD] immediates.
2400 (movti): Split CONST_POLY_INT moves into two halves.
2401 (add<mode>3): Accept aarch64_pluslong_or_poly_operand.
2402 Split additions that need a temporary here if the destination
2403 is the stack pointer.
2404 (*add<mode>3_aarch64): Handle ADDVL and ADDPL immediates.
2405 (*add<mode>3_poly_1): New instruction.
2406 (set_clobber_cc): New expander.
2408 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2410 * simplify-rtx.c (simplify_immed_subreg): Add an inner_bytes
2411 parameter and use it instead of GET_MODE_SIZE (innermode). Use
2412 inner_bytes * BITS_PER_UNIT instead of GET_MODE_BITSIZE (innermode).
2413 Use CEIL (inner_bytes, GET_MODE_UNIT_SIZE (innermode)) instead of
2414 GET_MODE_NUNITS (innermode). Also add a first_elem parameter.
2415 Change innermode from fixed_mode_size to machine_mode.
2416 (simplify_subreg): Update call accordingly. Handle a constant-sized
2417 subreg of a variable-length CONST_VECTOR.
2419 2018-01-13 Richard Sandiford <richard.sandiford@linaro.org>
2420 Alan Hayward <alan.hayward@arm.com>
2421 David Sherwood <david.sherwood@arm.com>
2423 * tree-ssa-address.c (mem_ref_valid_without_offset_p): New function.
2424 (add_offset_to_base): New function, split out from...
2425 (create_mem_ref): ...here. When handling a scale other than 1,
2426 check first whether the address is valid without the offset.
2427 Add it into the base if so, leaving the index and scale as-is.
2429 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2432 * config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin): Call
2433 fold_for_warn before checking if arg2 is INTEGER_CST.
2435 2018-01-12 Segher Boessenkool <segher@kernel.crashing.org>
2437 * config/rs6000/predicates.md (load_multiple_operation): Delete.
2438 (store_multiple_operation): Delete.
2439 * config/rs6000/rs6000-cpus.def (601): Remove MASK_STRING.
2440 * config/rs6000/rs6000-protos.h (rs6000_output_load_multiple): Delete.
2441 * config/rs6000/rs6000-string.c (expand_block_move): Delete everything
2442 guarded by TARGET_STRING.
2443 (rs6000_output_load_multiple): Delete.
2444 * config/rs6000/rs6000.c (rs6000_option_override_internal): Delete
2445 OPTION_MASK_STRING / TARGET_STRING handling.
2446 (print_operand) <'N', 'O'>: Add comment that these are unused now.
2447 (const rs6000_opt_masks) <"string">: Change mask to 0.
2448 * config/rs6000/rs6000.h (TARGET_DEFAULT): Remove MASK_STRING.
2449 (MASK_STRING): Delete.
2450 * config/rs6000/rs6000.md (*mov<mode>_string): Delete TARGET_STRING
2452 (load_multiple): Delete.
2459 (store_multiple): Delete.
2466 (movmemsi_8reg): Delete.
2467 (corresponding unnamed define_insn): Delete.
2468 (movmemsi_6reg): Delete.
2469 (corresponding unnamed define_insn): Delete.
2470 (movmemsi_4reg): Delete.
2471 (corresponding unnamed define_insn): Delete.
2472 (movmemsi_2reg): Delete.
2473 (corresponding unnamed define_insn): Delete.
2474 (movmemsi_1reg): Delete.
2475 (corresponding unnamed define_insn): Delete.
2476 * config/rs6000/rs6000.opt (mno-string): New.
2477 (mstring): Replace by deprecation warning stub.
2478 * doc/invoke.texi (RS/6000 and PowerPC Options): Delete -mstring.
2480 2018-01-12 Jakub Jelinek <jakub@redhat.com>
2482 * regrename.c (regrename_do_replace): If replacing the same
2483 reg multiple times, try to reuse last created gen_raw_REG.
2486 * bb-reorder.c (pass_partition_blocks::gate): In lto don't partition
2487 main to workaround a bug in GDB.
2489 2018-01-12 Tom de Vries <tom@codesourcery.com>
2492 * config.gcc (nvptx*-*-*): Set use_gcc_stdint=wrap.
2494 2018-01-12 Vladimir Makarov <vmakarov@redhat.com>
2496 PR rtl-optimization/80481
2497 * ira-color.c (get_cap_member): New function.
2498 (allocnos_conflict_by_live_ranges_p): Use it.
2499 (slot_coalesced_allocno_live_ranges_intersect_p): Add assert.
2500 (setup_slot_coalesced_allocno_live_ranges): Ditto.
2502 2018-01-12 Uros Bizjak <ubizjak@gmail.com>
2505 * config/alpha/alpha.md (*saddsi_1): New insn_ans_split pattern.
2506 (*saddl_se_1): Ditto.
2508 (*ssubl_se_1): Ditto.
2510 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2512 * tree-predcom.c (aff_combination_dr_offset): Use wi::to_poly_widest
2513 rather than wi::to_widest for DR_INITs.
2514 * tree-vect-data-refs.c (vect_find_same_alignment_drs): Use
2515 wi::to_poly_offset rather than wi::to_offset for DR_INIT.
2516 (vect_analyze_data_ref_accesses): Require both DR_INITs to be
2518 (vect_analyze_group_access_1): Note that here.
2520 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2522 * tree-vectorizer.c (get_vec_alignment_for_array_type): Handle
2523 polynomial type sizes.
2525 2018-01-12 Richard Sandiford <richard.sandiford@linaro.org>
2527 * gimplify.c (gimple_add_tmp_var_fn): Allow variables to have a
2528 poly_uint64 size, rather than requiring an unsigned HOST_WIDE_INT size.
2529 (gimple_add_tmp_var): Likewise.
2531 2018-01-12 Martin Liska <mliska@suse.cz>
2533 * gimple.c (gimple_alloc_counts): Use uint64_t instead of int.
2534 (gimple_alloc_sizes): Likewise.
2535 (dump_gimple_statistics): Use PRIu64 in printf format.
2536 * gimple.h: Change uint64_t to int.
2538 2018-01-12 Martin Liska <mliska@suse.cz>
2540 * tree-core.h: Use uint64_t instead of int.
2541 * tree.c (tree_node_counts): Likewise.
2542 (tree_node_sizes): Likewise.
2543 (dump_tree_statistics): Use PRIu64 in printf format.
2545 2018-01-12 Martin Liska <mliska@suse.cz>
2547 * Makefile.in: As qsort_chk is implemented in vec.c, add
2548 vec.o to linkage of gencfn-macros.
2549 * tree.c (build_new_poly_int_cst): Add CXX_MEM_STAT_INFO as it's
2550 passing the info to record_node_allocation_statistics.
2551 (test_vector_cst_patterns): Add CXX_MEM_STAT_INFO to declaration
2553 * ggc-common.c (struct ggc_usage): Add operator== and use
2554 it in operator< and compare function.
2555 * mem-stats.h (struct mem_usage): Likewise.
2556 * vec.c (struct vec_usage): Remove operator< and compare
2557 function. Can be simply inherited.
2559 2018-01-12 Martin Jambor <mjambor@suse.cz>
2562 * params.def: New parameter PARAM_AVOID_FMA_MAX_BITS.
2563 * tree-ssa-math-opts.c: Include domwalk.h.
2564 (convert_mult_to_fma_1): New function.
2565 (fma_transformation_info): New type.
2566 (fma_deferring_state): Likewise.
2567 (cancel_fma_deferring): New function.
2568 (result_of_phi): Likewise.
2569 (last_fma_candidate_feeds_initial_phi): Likewise.
2570 (convert_mult_to_fma): Added deferring logic, split actual
2571 transformation to convert_mult_to_fma_1.
2572 (math_opts_dom_walker): New type.
2573 (math_opts_dom_walker::after_dom_children): New method, body moved
2574 here from pass_optimize_widening_mul::execute, added deferring logic
2576 (pass_optimize_widening_mul::execute): Moved most of code to
2577 math_opts_dom_walker::after_dom_children.
2578 * config/i386/x86-tune.def (X86_TUNE_AVOID_128FMA_CHAINS): New.
2579 * config/i386/i386.c (ix86_option_override_internal): Added
2580 maybe_setting of PARAM_AVOID_FMA_MAX_BITS.
2582 2018-01-12 Richard Biener <rguenther@suse.de>
2585 * dwarf2out.c (gen_variable_die): Do not reset old_die for
2586 inline instance vars.
2588 2018-01-12 Oleg Endo <olegendo@gcc.gnu.org>
2591 * config/rx/rx.c (rx_is_restricted_memory_address):
2594 2018-01-12 Richard Biener <rguenther@suse.de>
2596 PR tree-optimization/80846
2597 * target.def (split_reduction): New target hook.
2598 * targhooks.c (default_split_reduction): New function.
2599 * targhooks.h (default_split_reduction): Declare.
2600 * tree-vect-loop.c (vect_create_epilog_for_reduction): If the
2601 target requests first reduce vectors by combining low and high
2603 * tree-vect-stmts.c (vect_gen_perm_mask_any): Adjust.
2604 (get_vectype_for_scalar_type_and_size): Export.
2605 * tree-vectorizer.h (get_vectype_for_scalar_type_and_size): Declare.
2606 * doc/tm.texi.in (TARGET_VECTORIZE_SPLIT_REDUCTION): Document.
2607 * doc/tm.texi: Regenerate.
2608 * config/i386/i386.c (ix86_split_reduction): Implement
2609 TARGET_VECTORIZE_SPLIT_REDUCTION.
2611 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2614 * config/sparc/sparc.h (PIC_OFFSET_TABLE_REGNUM): Set to INVALID_REGNUM
2615 in PIC mode except for TARGET_VXWORKS_RTP.
2616 * config/sparc/sparc.c: Include cfgrtl.h.
2617 (TARGET_INIT_PIC_REG): Define.
2618 (TARGET_USE_PSEUDO_PIC_REG): Likewise.
2619 (sparc_pic_register_p): New predicate.
2620 (sparc_legitimate_address_p): Use it.
2621 (sparc_legitimize_pic_address): Likewise.
2622 (sparc_delegitimize_address): Likewise.
2623 (sparc_mode_dependent_address_p): Likewise.
2624 (gen_load_pcrel_sym): Remove 4th parameter.
2625 (load_got_register): Adjust call to above. Remove obsolete stuff.
2626 (sparc_expand_prologue): Do not call load_got_register here.
2627 (sparc_flat_expand_prologue): Likewise.
2628 (sparc_output_mi_thunk): Set the pic_offset_table_rtx object.
2629 (sparc_use_pseudo_pic_reg): New function.
2630 (sparc_init_pic_reg): Likewise.
2631 * config/sparc/sparc.md (vxworks_load_got): Set the GOT register.
2632 (builtin_setjmp_receiver): Enable only for TARGET_VXWORKS_RTP.
2634 2018-01-12 Christophe Lyon <christophe.lyon@linaro.org>
2636 * doc/sourcebuild.texi (Effective-Target Keywords, Other attributes):
2637 Add item for branch_cost.
2639 2018-01-12 Eric Botcazou <ebotcazou@adacore.com>
2641 PR rtl-optimization/83565
2642 * rtlanal.c (nonzero_bits1): On WORD_REGISTER_OPERATIONS machines, do
2643 not extend the result to a larger mode for rotate operations.
2644 (num_sign_bit_copies1): Likewise.
2646 2018-01-12 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
2649 * config/sol2.h (STARTFILE_ARCH_SPEC): Don't use with -shared or
2651 Use values-Xc.o for -pedantic.
2652 Link with values-xpg4.o for C90, values-xpg6.o otherwise.
2654 2018-01-12 Martin Liska <mliska@suse.cz>
2657 * ipa-devirt.c (final_warning_record::grow_type_warnings):
2659 (possible_polymorphic_call_targets): Use it.
2660 (ipa_devirt): Likewise.
2662 2018-01-12 Martin Liska <mliska@suse.cz>
2664 * profile-count.h (enum profile_quality): Use 0 as invalid
2665 enum value of profile_quality.
2667 2018-01-12 Chung-Ju Wu <jasonwucj@gmail.com>
2669 * doc/invoke.texi (NDS32 Options): Add -mext-perf, -mext-perf2 and
2670 -mext-string options.
2672 2018-01-12 Richard Biener <rguenther@suse.de>
2674 * lto-streamer-out.c (DFS::DFS_write_tree_body): Process
2675 DECL_DEBUG_EXPR conditional on DECL_HAS_DEBUG_EXPR_P.
2676 * tree-streamer-in.c (lto_input_ts_decl_common_tree_pointers):
2678 * tree-streamer-out.c (write_ts_decl_common_tree_pointers): Likewise.
2680 2018-01-11 Michael Meissner <meissner@linux.vnet.ibm.com>
2682 * configure.ac (--with-long-double-format): Add support for the
2683 configuration option to change the default long double format on
2685 * config.gcc (powerpc*-linux*-*): Likewise.
2686 * configure: Regenerate.
2687 * config/rs6000/rs6000-c.c (rs6000_cpu_cpp_builtins): If long
2688 double is IEEE, define __KC__ and __KF__ to allow floatn.h to be
2689 used without modification.
2691 2018-01-11 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
2693 * config/rs6000/rs6000-builtin.def (BU_P7_MISC_X): New #define.
2694 (SPEC_BARRIER): New instantiation of BU_P7_MISC_X.
2695 * config/rs6000/rs6000.c (rs6000_expand_builtin): Handle
2696 MISC_BUILTIN_SPEC_BARRIER.
2697 (rs6000_init_builtins): Likewise.
2698 * config/rs6000/rs6000.md (UNSPECV_SPEC_BARRIER): New UNSPECV
2700 (speculation_barrier): New define_insn.
2701 * doc/extend.texi: Document __builtin_speculation_barrier.
2703 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2706 * config/i386/i386.c (ix86_expand_vector_init_one_nonzero): If one_var
2707 is 0, for V{8,16}S[IF] and V[48]D[IF]mode use gen_vec_set<mode>_0.
2708 * config/i386/sse.md (VI8_AVX_AVX512F, VI4F_256_512): New mode
2710 (ssescalarmodesuffix): Add 512-bit vectors. Use "d" or "q" for
2711 integral modes instead of "ss" and "sd".
2712 (vec_set<mode>_0): New define_insns for 256-bit and 512-bit
2713 vectors with 32-bit and 64-bit elements.
2714 (vecdupssescalarmodesuffix): New mode attribute.
2715 (vec_dup<mode>): Use it.
2717 2018-01-11 H.J. Lu <hongjiu.lu@intel.com>
2720 * config/i386/i386.c (ix86_compute_frame_layout): Align stack
2721 frame if argument is passed on stack.
2723 2018-01-11 Jakub Jelinek <jakub@redhat.com>
2726 * ree.c (combine_reaching_defs): Optimize also
2727 reg2=exp; reg1=reg2; reg2=any_extend(reg1); into
2728 reg2=any_extend(exp); reg1=reg2;, formatting fix.
2730 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2733 * gimple-ssa-isolate-paths.c (isolate_path): Fix profile update.
2735 2018-01-11 Jan Hubicka <hubicka@ucw.cz>
2738 * tree-inline.c (copy_cfg_body): Adjust num&den for scaling
2739 after they are computed.
2741 2018-01-11 Bin Cheng <bin.cheng@arm.com>
2743 PR tree-optimization/83695
2744 * gimple-loop-linterchange.cc
2745 (tree_loop_interchange::interchange_loops): Call scev_reset_htab to
2746 reset cached scev information after interchange.
2747 (pass_linterchange::execute): Remove call to scev_reset_htab.
2749 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2751 * config/arm/arm_neon.h (vfmlal_lane_low_u32, vfmlal_lane_high_u32,
2752 vfmlalq_laneq_low_u32, vfmlalq_lane_low_u32, vfmlal_laneq_low_u32,
2753 vfmlalq_laneq_high_u32, vfmlalq_lane_high_u32, vfmlal_laneq_high_u32,
2754 vfmlsl_lane_low_u32, vfmlsl_lane_high_u32, vfmlslq_laneq_low_u32,
2755 vfmlslq_lane_low_u32, vfmlsl_laneq_low_u32, vfmlslq_laneq_high_u32,
2756 vfmlslq_lane_high_u32, vfmlsl_laneq_high_u32): Define.
2757 * config/arm/arm_neon_builtins.def (vfmal_lane_low,
2758 vfmal_lane_lowv4hf, vfmal_lane_lowv8hf, vfmal_lane_high,
2759 vfmal_lane_highv4hf, vfmal_lane_highv8hf, vfmsl_lane_low,
2760 vfmsl_lane_lowv4hf, vfmsl_lane_lowv8hf, vfmsl_lane_high,
2761 vfmsl_lane_highv4hf, vfmsl_lane_highv8hf): New sets of builtins.
2762 * config/arm/iterators.md (VFMLSEL2, vfmlsel2): New mode attributes.
2763 (V_lane_reg): Likewise.
2764 * config/arm/neon.md (neon_vfm<vfml_op>l_lane_<vfml_half><VCVTF:mode>):
2766 (neon_vfm<vfml_op>l_lane_<vfml_half><vfmlsel2><mode>): Likewise.
2767 (vfmal_lane_low<mode>_intrinsic,
2768 vfmal_lane_low<vfmlsel2><mode>_intrinsic,
2769 vfmal_lane_high<vfmlsel2><mode>_intrinsic,
2770 vfmal_lane_high<mode>_intrinsic, vfmsl_lane_low<mode>_intrinsic,
2771 vfmsl_lane_low<vfmlsel2><mode>_intrinsic,
2772 vfmsl_lane_high<vfmlsel2><mode>_intrinsic,
2773 vfmsl_lane_high<mode>_intrinsic): New define_insns.
2775 2018-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2777 * config/arm/arm-cpus.in (fp16fml): New feature.
2778 (ALL_SIMD): Add fp16fml.
2779 (armv8.2-a): Add fp16fml as an option.
2780 (armv8.3-a): Likewise.
2781 (armv8.4-a): Add fp16fml as part of fp16.
2782 * config/arm/arm.h (TARGET_FP16FML): Define.
2783 * config/arm/arm-c.c (arm_cpu_builtins): Define __ARM_FEATURE_FP16_FML
2785 * config/arm/arm-modes.def (V2HF): Define.
2786 * config/arm/arm_neon.h (vfmlal_low_u32, vfmlsl_low_u32,
2787 vfmlal_high_u32, vfmlsl_high_u32, vfmlalq_low_u32,
2788 vfmlslq_low_u32, vfmlalq_high_u32, vfmlslq_high_u32): Define.
2789 * config/arm/arm_neon_builtins.def (vfmal_low, vfmal_high,
2790 vfmsl_low, vfmsl_high): New set of builtins.
2791 * config/arm/iterators.md (PLUSMINUS): New code iterator.
2792 (vfml_op): New code attribute.
2793 (VFMLHALVES): New int iterator.
2794 (VFML, VFMLSEL): New mode attributes.
2795 (V_reg): Define mapping for V2HF.
2796 (V_hi, V_lo): New mode attributes.
2797 (VF_constraint): Likewise.
2798 (vfml_half, vfml_half_selector): New int attributes.
2799 * config/arm/neon.md (neon_vfm<vfml_op>l_<vfml_half><mode>): New
2801 (vfmal_low<mode>_intrinsic, vfmsl_high<mode>_intrinsic,
2802 vfmal_high<mode>_intrinsic, vfmsl_low<mode>_intrinsic):
2804 * config/arm/t-arm-elf (v8_fps): Add fp16fml.
2805 * config/arm/t-multilib (v8_2_a_simd_variants): Add fp16fml.
2806 * config/arm/unspecs.md (UNSPEC_VFML_LO, UNSPEC_VFML_HI): New unspecs.
2807 * doc/invoke.texi (ARM Options): Document fp16fml. Update armv8.4-a
2809 * doc/sourcebuild.texi (arm_fp16fml_neon_ok, arm_fp16fml_neon):
2810 Document new effective target and option set.
2812 2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
2814 * config/arm/arm-cpus.in (armv8_4): New feature.
2815 (ARMv8_4a): New fgroup.
2816 (armv8.4-a): New arch.
2817 * config/arm/arm-tables.opt: Regenerate.
2818 * config/arm/t-aprofile: Add matching rules for -march=armv8.4-a.
2819 * config/arm/t-arm-elf (all_v8_archs): Add armv8.4-a.
2820 * config/arm/t-multilib (v8_4_a_simd_variants): New variable.
2821 Add matching rules for -march=armv8.4-a and extensions.
2822 * doc/invoke.texi (ARM Options): Document -march=armv8.4-a.
2824 2018-01-11 Oleg Endo <olegendo@gcc.gnu.org>
2827 * config/rx/rx.md (BW): New mode attribute.
2828 (sync_lock_test_and_setsi): Add mode suffix to insn output.
2830 2018-01-11 Richard Biener <rguenther@suse.de>
2832 PR tree-optimization/83435
2833 * graphite.c (canonicalize_loop_form): Ignore fake loop exit edges.
2834 * graphite-scop-detection.c (scop_detection::get_sese): Likewise.
2835 * tree-vrp.c (add_assert_info): Drop TREE_OVERFLOW if they appear.
2837 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2838 Alan Hayward <alan.hayward@arm.com>
2839 David Sherwood <david.sherwood@arm.com>
2841 * config/aarch64/aarch64.c (aarch64_address_info): Add a const_offset
2843 (aarch64_classify_address): Initialize it. Track polynomial offsets.
2844 (aarch64_print_address_internal): Use it to check for a zero offset.
2846 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2847 Alan Hayward <alan.hayward@arm.com>
2848 David Sherwood <david.sherwood@arm.com>
2850 * config/aarch64/aarch64-modes.def (NUM_POLY_INT_COEFFS): Set to 2.
2851 * config/aarch64/aarch64-protos.h (aarch64_initial_elimination_offset):
2852 Return a poly_int64 rather than a HOST_WIDE_INT.
2853 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a poly_int64
2854 rather than a HOST_WIDE_INT.
2855 * config/aarch64/aarch64.h (aarch64_frame): Protect with
2856 HAVE_POLY_INT_H rather than HOST_WIDE_INT. Change locals_offset,
2857 hard_fp_offset, frame_size, initial_adjust, callee_offset and
2858 final_offset from HOST_WIDE_INT to poly_int64.
2859 * config/aarch64/aarch64-builtins.c (aarch64_simd_expand_args): Use
2860 to_constant when getting the number of units in an Advanced SIMD
2862 (aarch64_builtin_vectorized_function): Check for a constant number
2864 * config/aarch64/aarch64-simd.md (mov<mode>): Handle polynomial
2866 (aarch64_ld<VSTRUCT:nregs>_lane<VALLDIF:mode>): Use the nunits
2867 attribute instead of GET_MODE_NUNITS.
2868 * config/aarch64/aarch64.c (aarch64_hard_regno_nregs)
2869 (aarch64_class_max_nregs): Use the constant_lowest_bound of the
2870 GET_MODE_SIZE for fixed-size registers.
2871 (aarch64_const_vec_all_same_in_range_p): Use const_vec_duplicate_p.
2872 (aarch64_hard_regno_call_part_clobbered, aarch64_classify_index)
2873 (aarch64_mode_valid_for_sched_fusion_p, aarch64_classify_address)
2874 (aarch64_legitimize_address_displacement, aarch64_secondary_reload)
2875 (aarch64_print_operand, aarch64_print_address_internal)
2876 (aarch64_address_cost, aarch64_rtx_costs, aarch64_register_move_cost)
2877 (aarch64_short_vector_p, aapcs_vfp_sub_candidate)
2878 (aarch64_simd_attr_length_rglist, aarch64_operands_ok_for_ldpstp):
2879 Handle polynomial GET_MODE_SIZE.
2880 (aarch64_hard_regno_caller_save_mode): Likewise. Return modes
2881 wider than SImode without modification.
2882 (tls_symbolic_operand_type): Use strip_offset instead of split_const.
2883 (aarch64_pass_by_reference, aarch64_layout_arg, aarch64_pad_reg_upward)
2884 (aarch64_gimplify_va_arg_expr): Assert that we don't yet handle
2885 passing and returning SVE modes.
2886 (aarch64_function_value, aarch64_layout_arg): Use gen_int_mode
2887 rather than GEN_INT.
2888 (aarch64_emit_probe_stack_range): Take the size as a poly_int64
2889 rather than a HOST_WIDE_INT, but call sorry if it isn't constant.
2890 (aarch64_allocate_and_probe_stack_space): Likewise.
2891 (aarch64_layout_frame): Cope with polynomial offsets.
2892 (aarch64_save_callee_saves, aarch64_restore_callee_saves): Take the
2893 start_offset as a poly_int64 rather than a HOST_WIDE_INT. Track
2895 (offset_9bit_signed_unscaled_p, offset_12bit_unsigned_scaled_p)
2896 (aarch64_offset_7bit_signed_scaled_p): Take the offset as a
2897 poly_int64 rather than a HOST_WIDE_INT.
2898 (aarch64_get_separate_components, aarch64_process_components)
2899 (aarch64_expand_prologue, aarch64_expand_epilogue)
2900 (aarch64_use_return_insn_p): Handle polynomial frame offsets.
2901 (aarch64_anchor_offset): New function, split out from...
2902 (aarch64_legitimize_address): ...here.
2903 (aarch64_builtin_vectorization_cost): Handle polynomial
2904 TYPE_VECTOR_SUBPARTS.
2905 (aarch64_simd_check_vect_par_cnst_half): Handle polynomial
2907 (aarch64_simd_make_constant, aarch64_expand_vector_init): Get the
2908 number of elements from the PARALLEL rather than the mode.
2909 (aarch64_shift_truncation_mask): Use GET_MODE_UNIT_BITSIZE
2910 rather than GET_MODE_BITSIZE.
2911 (aarch64_evpc_trn, aarch64_evpc_uzp, aarch64_evpc_ext)
2912 (aarch64_evpc_rev, aarch64_evpc_dup, aarch64_evpc_zip)
2913 (aarch64_expand_vec_perm_const_1): Handle polynomial
2914 d->perm.length () and d->perm elements.
2915 (aarch64_evpc_tbl): Likewise. Use nelt rather than GET_MODE_NUNITS.
2916 Apply to_constant to d->perm elements.
2917 (aarch64_simd_valid_immediate, aarch64_vec_fpconst_pow_of_2): Handle
2918 polynomial CONST_VECTOR_NUNITS.
2919 (aarch64_move_pointer): Take amount as a poly_int64 rather
2921 (aarch64_progress_pointer): Avoid temporary variable.
2922 * config/aarch64/aarch64.md (aarch64_<crc_variant>): Use
2923 the mode attribute instead of GET_MODE.
2925 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2926 Alan Hayward <alan.hayward@arm.com>
2927 David Sherwood <david.sherwood@arm.com>
2929 * config/aarch64/aarch64.c (aarch64_force_temporary): Assert that
2930 x exists before using it.
2931 (aarch64_add_constant_internal): Rename to...
2932 (aarch64_add_offset_1): ...this. Replace regnum with separate
2933 src and dest rtxes. Handle the case in which they're different,
2934 including when the offset is zero. Replace scratchreg with an rtx.
2935 Use 2 additions if there is no spare register into which we can
2936 move a 16-bit constant.
2937 (aarch64_add_constant): Delete.
2938 (aarch64_add_offset): Replace reg with separate src and dest
2939 rtxes. Take a poly_int64 offset instead of a HOST_WIDE_INT.
2940 Use aarch64_add_offset_1.
2941 (aarch64_add_sp, aarch64_sub_sp): Take the scratch register as
2942 an rtx rather than an int. Take the delta as a poly_int64
2943 rather than a HOST_WIDE_INT. Use aarch64_add_offset.
2944 (aarch64_expand_mov_immediate): Update uses of aarch64_add_offset.
2945 (aarch64_expand_prologue): Update calls to aarch64_sub_sp,
2946 aarch64_allocate_and_probe_stack_space and aarch64_add_offset.
2947 (aarch64_expand_epilogue): Update calls to aarch64_add_offset
2949 (aarch64_output_mi_thunk): Use aarch64_add_offset rather than
2950 aarch64_add_constant.
2952 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2954 * config/aarch64/aarch64.c (aarch64_reinterpret_float_as_int):
2955 Use scalar_float_mode.
2957 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2959 * config/aarch64/aarch64-simd.md
2960 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): Avoid GET_MODE_NUNITS.
2961 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Likewise.
2962 (aarch64_fml<f16mac1>l_lane_lowv2sf): Likewise.
2963 (aarch64_fml<f16mac1>l_lane_highv2sf): Likewise.
2964 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Likewise.
2965 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Likewise.
2966 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Likewise.
2967 (aarch64_fml<f16mac1>l_laneq_highv2sf): Likewise.
2968 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Likewise.
2969 (aarch64_fml<f16mac1>lq_lane_highv4sf): Likewise.
2971 2018-01-11 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
2974 * config/arm/arm.c (arm_declare_function_name): Set arch_to_print if
2975 targ_options->x_arm_arch_string is non NULL.
2977 2018-01-11 Tamar Christina <tamar.christina@arm.com>
2979 * config/aarch64/aarch64.h
2980 (AARCH64_FL_FOR_ARCH8_4): Add AARCH64_FL_DOTPROD.
2982 2018-01-11 Sudakshina Das <sudi.das@arm.com>
2985 * expmed.c (emit_store_flag_force): Swap if const op0
2986 and change VOIDmode to mode of op0.
2988 2018-01-11 Richard Sandiford <richard.sandiford@linaro.org>
2990 PR rtl-optimization/83761
2991 * caller-save.c (replace_reg_with_saved_mem): Pass bits rather
2992 than bytes to mode_for_size.
2994 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
2997 * gfortran.fortran-torture/compile/pr83189.f90: New testcase.
2998 * tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Handle zero
3001 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3004 * cfgrtl.c (rtl_verify_edges): Only verify fixability of partition
3005 when in layout mode.
3006 (cfg_layout_finalize): Do not verify cfg before we are out of layout.
3007 * cfgcleanup.c (try_optimize_cfg): Only verify flow info when doing
3010 2018-01-10 Michael Collison <michael.collison@arm.com>
3012 * config/aarch64/aarch64-modes.def (V2HF): New VECTOR_MODE.
3013 * config/aarch64/aarch64-option-extension.def: Add
3014 AARCH64_OPT_EXTENSION of 'fp16fml'.
3015 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3016 (__ARM_FEATURE_FP16_FML): Define if TARGET_F16FML is true.
3017 * config/aarch64/predicates.md (aarch64_lane_imm3): New predicate.
3018 * config/aarch64/constraints.md (Ui7): New constraint.
3019 * config/aarch64/iterators.md (VFMLA_W): New mode iterator.
3020 (VFMLA_SEL_W): Ditto.
3023 (VFMLA16_LOW): New int iterator.
3024 (VFMLA16_HIGH): Ditto.
3025 (UNSPEC_FMLAL): New unspec.
3026 (UNSPEC_FMLSL): Ditto.
3027 (UNSPEC_FMLAL2): Ditto.
3028 (UNSPEC_FMLSL2): Ditto.
3029 (f16mac): New code attribute.
3030 * config/aarch64/aarch64-simd-builtins.def
3031 (aarch64_fmlal_lowv2sf): Ditto.
3032 (aarch64_fmlsl_lowv2sf): Ditto.
3033 (aarch64_fmlalq_lowv4sf): Ditto.
3034 (aarch64_fmlslq_lowv4sf): Ditto.
3035 (aarch64_fmlal_highv2sf): Ditto.
3036 (aarch64_fmlsl_highv2sf): Ditto.
3037 (aarch64_fmlalq_highv4sf): Ditto.
3038 (aarch64_fmlslq_highv4sf): Ditto.
3039 (aarch64_fmlal_lane_lowv2sf): Ditto.
3040 (aarch64_fmlsl_lane_lowv2sf): Ditto.
3041 (aarch64_fmlal_laneq_lowv2sf): Ditto.
3042 (aarch64_fmlsl_laneq_lowv2sf): Ditto.
3043 (aarch64_fmlalq_lane_lowv4sf): Ditto.
3044 (aarch64_fmlsl_lane_lowv4sf): Ditto.
3045 (aarch64_fmlalq_laneq_lowv4sf): Ditto.
3046 (aarch64_fmlsl_laneq_lowv4sf): Ditto.
3047 (aarch64_fmlal_lane_highv2sf): Ditto.
3048 (aarch64_fmlsl_lane_highv2sf): Ditto.
3049 (aarch64_fmlal_laneq_highv2sf): Ditto.
3050 (aarch64_fmlsl_laneq_highv2sf): Ditto.
3051 (aarch64_fmlalq_lane_highv4sf): Ditto.
3052 (aarch64_fmlsl_lane_highv4sf): Ditto.
3053 (aarch64_fmlalq_laneq_highv4sf): Ditto.
3054 (aarch64_fmlsl_laneq_highv4sf): Ditto.
3055 * config/aarch64/aarch64-simd.md:
3056 (aarch64_fml<f16mac1>l<f16quad>_low<mode>): New pattern.
3057 (aarch64_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3058 (aarch64_simd_fml<f16mac1>l<f16quad>_low<mode>): Ditto.
3059 (aarch64_simd_fml<f16mac1>l<f16quad>_high<mode>): Ditto.
3060 (aarch64_fml<f16mac1>l_lane_lowv2sf): Ditto.
3061 (aarch64_fml<f16mac1>l_lane_highv2sf): Ditto.
3062 (aarch64_simd_fml<f16mac>l_lane_lowv2sf): Ditto.
3063 (aarch64_simd_fml<f16mac>l_lane_highv2sf): Ditto.
3064 (aarch64_fml<f16mac1>lq_laneq_lowv4sf): Ditto.
3065 (aarch64_fml<f16mac1>lq_laneq_highv4sf): Ditto.
3066 (aarch64_simd_fml<f16mac>lq_laneq_lowv4sf): Ditto.
3067 (aarch64_simd_fml<f16mac>lq_laneq_highv4sf): Ditto.
3068 (aarch64_fml<f16mac1>l_laneq_lowv2sf): Ditto.
3069 (aarch64_fml<f16mac1>l_laneq_highv2sf): Ditto.
3070 (aarch64_simd_fml<f16mac>l_laneq_lowv2sf): Ditto.
3071 (aarch64_simd_fml<f16mac>l_laneq_highv2sf): Ditto.
3072 (aarch64_fml<f16mac1>lq_lane_lowv4sf): Ditto.
3073 (aarch64_fml<f16mac1>lq_lane_highv4sf): Ditto.
3074 (aarch64_simd_fml<f16mac>lq_lane_lowv4sf): Ditto.
3075 (aarch64_simd_fml<f16mac>lq_lane_highv4sf): Ditto.
3076 * config/aarch64/arm_neon.h (vfmlal_low_u32): New intrinsic.
3077 (vfmlsl_low_u32): Ditto.
3078 (vfmlalq_low_u32): Ditto.
3079 (vfmlslq_low_u32): Ditto.
3080 (vfmlal_high_u32): Ditto.
3081 (vfmlsl_high_u32): Ditto.
3082 (vfmlalq_high_u32): Ditto.
3083 (vfmlslq_high_u32): Ditto.
3084 (vfmlal_lane_low_u32): Ditto.
3085 (vfmlsl_lane_low_u32): Ditto.
3086 (vfmlal_laneq_low_u32): Ditto.
3087 (vfmlsl_laneq_low_u32): Ditto.
3088 (vfmlalq_lane_low_u32): Ditto.
3089 (vfmlslq_lane_low_u32): Ditto.
3090 (vfmlalq_laneq_low_u32): Ditto.
3091 (vfmlslq_laneq_low_u32): Ditto.
3092 (vfmlal_lane_high_u32): Ditto.
3093 (vfmlsl_lane_high_u32): Ditto.
3094 (vfmlal_laneq_high_u32): Ditto.
3095 (vfmlsl_laneq_high_u32): Ditto.
3096 (vfmlalq_lane_high_u32): Ditto.
3097 (vfmlslq_lane_high_u32): Ditto.
3098 (vfmlalq_laneq_high_u32): Ditto.
3099 (vfmlslq_laneq_high_u32): Ditto.
3100 * config/aarch64/aarch64.h (AARCH64_FL_F16SML): New flag.
3101 (AARCH64_FL_FOR_ARCH8_4): New.
3102 (AARCH64_ISA_F16FML): New ISA flag.
3103 (TARGET_F16FML): New feature flag for fp16fml.
3104 (doc/invoke.texi): Document new fp16fml option.
3106 2018-01-10 Michael Collison <michael.collison@arm.com>
3108 * config/aarch64/aarch64-builtins.c:
3109 (aarch64_types_ternopu_imm_qualifiers, TYPES_TERNOPUI): New.
3110 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3111 (__ARM_FEATURE_SHA3): Define if TARGET_SHA3 is true.
3112 * config/aarch64/aarch64.h (AARCH64_FL_SHA3): New flags.
3113 (AARCH64_ISA_SHA3): New ISA flag.
3114 (TARGET_SHA3): New feature flag for sha3.
3115 * config/aarch64/iterators.md (sha512_op): New int attribute.
3116 (CRYPTO_SHA512): New int iterator.
3117 (UNSPEC_SHA512H): New unspec.
3118 (UNSPEC_SHA512H2): Ditto.
3119 (UNSPEC_SHA512SU0): Ditto.
3120 (UNSPEC_SHA512SU1): Ditto.
3121 * config/aarch64/aarch64-simd-builtins.def
3122 (aarch64_crypto_sha512hqv2di): New builtin.
3123 (aarch64_crypto_sha512h2qv2di): Ditto.
3124 (aarch64_crypto_sha512su0qv2di): Ditto.
3125 (aarch64_crypto_sha512su1qv2di): Ditto.
3126 (aarch64_eor3qv8hi): Ditto.
3127 (aarch64_rax1qv2di): Ditto.
3128 (aarch64_xarqv2di): Ditto.
3129 (aarch64_bcaxqv8hi): Ditto.
3130 * config/aarch64/aarch64-simd.md:
3131 (aarch64_crypto_sha512h<sha512_op>qv2di): New pattern.
3132 (aarch64_crypto_sha512su0qv2di): Ditto.
3133 (aarch64_crypto_sha512su1qv2di): Ditto.
3134 (aarch64_eor3qv8hi): Ditto.
3135 (aarch64_rax1qv2di): Ditto.
3136 (aarch64_xarqv2di): Ditto.
3137 (aarch64_bcaxqv8hi): Ditto.
3138 * config/aarch64/arm_neon.h (vsha512hq_u64): New intrinsic.
3139 (vsha512h2q_u64): Ditto.
3140 (vsha512su0q_u64): Ditto.
3141 (vsha512su1q_u64): Ditto.
3142 (veor3q_u16): Ditto.
3143 (vrax1q_u64): Ditto.
3145 (vbcaxq_u16): Ditto.
3146 * config/arm/types.md (crypto_sha512): New type attribute.
3147 (crypto_sha3): Ditto.
3148 (doc/invoke.texi): Document new sha3 option.
3150 2018-01-10 Michael Collison <michael.collison@arm.com>
3152 * config/aarch64/aarch64-builtins.c:
3153 (aarch64_types_quadopu_imm_qualifiers, TYPES_QUADOPUI): New.
3154 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3155 (__ARM_FEATURE_SM3): Define if TARGET_SM4 is true.
3156 (__ARM_FEATURE_SM4): Define if TARGET_SM4 is true.
3157 * config/aarch64/aarch64.h (AARCH64_FL_SM4): New flags.
3158 (AARCH64_ISA_SM4): New ISA flag.
3159 (TARGET_SM4): New feature flag for sm4.
3160 * config/aarch64/aarch64-simd-builtins.def
3161 (aarch64_sm3ss1qv4si): Ditto.
3162 (aarch64_sm3tt1aq4si): Ditto.
3163 (aarch64_sm3tt1bq4si): Ditto.
3164 (aarch64_sm3tt2aq4si): Ditto.
3165 (aarch64_sm3tt2bq4si): Ditto.
3166 (aarch64_sm3partw1qv4si): Ditto.
3167 (aarch64_sm3partw2qv4si): Ditto.
3168 (aarch64_sm4eqv4si): Ditto.
3169 (aarch64_sm4ekeyqv4si): Ditto.
3170 * config/aarch64/aarch64-simd.md:
3171 (aarch64_sm3ss1qv4si): Ditto.
3172 (aarch64_sm3tt<sm3tt_op>qv4si): Ditto.
3173 (aarch64_sm3partw<sm3part_op>qv4si): Ditto.
3174 (aarch64_sm4eqv4si): Ditto.
3175 (aarch64_sm4ekeyqv4si): Ditto.
3176 * config/aarch64/iterators.md (sm3tt_op): New int iterator.
3177 (sm3part_op): Ditto.
3178 (CRYPTO_SM3TT): Ditto.
3179 (CRYPTO_SM3PART): Ditto.
3180 (UNSPEC_SM3SS1): New unspec.
3181 (UNSPEC_SM3TT1A): Ditto.
3182 (UNSPEC_SM3TT1B): Ditto.
3183 (UNSPEC_SM3TT2A): Ditto.
3184 (UNSPEC_SM3TT2B): Ditto.
3185 (UNSPEC_SM3PARTW1): Ditto.
3186 (UNSPEC_SM3PARTW2): Ditto.
3187 (UNSPEC_SM4E): Ditto.
3188 (UNSPEC_SM4EKEY): Ditto.
3189 * config/aarch64/constraints.md (Ui2): New constraint.
3190 * config/aarch64/predicates.md (aarch64_imm2): New predicate.
3191 * config/arm/types.md (crypto_sm3): New type attribute.
3192 (crypto_sm4): Ditto.
3193 * config/aarch64/arm_neon.h (vsm3ss1q_u32): New intrinsic.
3194 (vsm3tt1aq_u32): Ditto.
3195 (vsm3tt1bq_u32): Ditto.
3196 (vsm3tt2aq_u32): Ditto.
3197 (vsm3tt2bq_u32): Ditto.
3198 (vsm3partw1q_u32): Ditto.
3199 (vsm3partw2q_u32): Ditto.
3200 (vsm4eq_u32): Ditto.
3201 (vsm4ekeyq_u32): Ditto.
3202 (doc/invoke.texi): Document new sm4 option.
3204 2018-01-10 Michael Collison <michael.collison@arm.com>
3206 * config/aarch64/aarch64-arches.def (armv8.4-a): New architecture.
3207 * config/aarch64/aarch64.h (AARCH64_ISA_V8_4): New ISA flag.
3208 (AARCH64_FL_FOR_ARCH8_4): New.
3209 (AARCH64_FL_V8_4): New flag.
3210 (doc/invoke.texi): Document new armv8.4-a option.
3212 2018-01-10 Michael Collison <michael.collison@arm.com>
3214 * config/aarch64/aarch64-c.c (aarch64_update_cpp_builtins):
3215 (__ARM_FEATURE_AES): Define if TARGET_AES is true.
3216 (__ARM_FEATURE_SHA2): Define if TARGET_SHA2 is true.
3217 * config/aarch64/aarch64-option-extension.def: Add
3218 AARCH64_OPT_EXTENSION of 'sha2'.
3219 (aes): Add AARCH64_OPT_EXTENSION of 'aes'.
3220 (crypto): Disable sha2 and aes if crypto disabled.
3221 (crypto): Enable aes and sha2 if enabled.
3222 (simd): Disable sha2 and aes if simd disabled.
3223 * config/aarch64/aarch64.h (AARCH64_FL_AES, AARCH64_FL_SHA2):
3225 (AARCH64_ISA_AES, AARCH64_ISA_SHA2): New ISA flags.
3226 (TARGET_SHA2): New feature flag for sha2.
3227 (TARGET_AES): New feature flag for aes.
3228 * config/aarch64/aarch64-simd.md:
3229 (aarch64_crypto_aes<aes_op>v16qi): Make pattern
3230 conditional on TARGET_AES.
3231 (aarch64_crypto_aes<aesmc_op>v16qi): Ditto.
3232 (aarch64_crypto_sha1hsi): Make pattern conditional
3234 (aarch64_crypto_sha1hv4si): Ditto.
3235 (aarch64_be_crypto_sha1hv4si): Ditto.
3236 (aarch64_crypto_sha1su1v4si): Ditto.
3237 (aarch64_crypto_sha1<sha1_op>v4si): Ditto.
3238 (aarch64_crypto_sha1su0v4si): Ditto.
3239 (aarch64_crypto_sha256h<sha256_op>v4si): Ditto.
3240 (aarch64_crypto_sha256su0v4si): Ditto.
3241 (aarch64_crypto_sha256su1v4si): Ditto.
3242 (doc/invoke.texi): Document new aes and sha2 options.
3244 2018-01-10 Martin Sebor <msebor@redhat.com>
3246 PR tree-optimization/83781
3247 * gimple-fold.c (get_range_strlen): Avoid treating arrays of pointers
3250 2018-01-11 Martin Sebor <msebor@gmail.com>
3251 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
3253 PR tree-optimization/83501
3254 PR tree-optimization/81703
3256 * tree-ssa-strlen.c (get_string_cst): Rename...
3257 (get_string_len): ...to this. Handle global constants.
3258 (handle_char_store): Adjust.
3260 2018-01-10 Kito Cheng <kito.cheng@gmail.com>
3261 Jim Wilson <jimw@sifive.com>
3263 * config/riscv/riscv-protos.h (riscv_output_return): New.
3264 * config/riscv/riscv.c (struct machine_function): New naked_p field.
3265 (riscv_attribute_table, riscv_output_return),
3266 (riscv_handle_fndecl_attribute, riscv_naked_function_p),
3267 (riscv_allocate_stack_slots_for_args, riscv_warn_func_return): New.
3268 (riscv_compute_frame_info): Only compute frame->mask if not a naked
3270 (riscv_expand_prologue): Add early return for naked function.
3271 (riscv_expand_epilogue): Likewise.
3272 (riscv_function_ok_for_sibcall): Return false for naked function.
3273 (riscv_set_current_function): New.
3274 (TARGET_SET_CURRENT_FUNCTION, TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS),
3275 (TARGET_ATTRIBUTE_TABLE, TARGET_WARN_FUNC_RETURN): New.
3276 * config/riscv/riscv.md (simple_return): Call riscv_output_return.
3277 * doc/extend.texi (RISC-V Function Attributes): New.
3279 2018-01-10 Michael Meissner <meissner@linux.vnet.ibm.com>
3281 * config/rs6000/rs6000.c (is_complex_IBM_long_double): Explicitly
3282 check for 128-bit long double before checking TCmode.
3283 * config/rs6000/rs6000.h (FLOAT128_IEEE_P): Explicitly check for
3284 128-bit long doubles before checking TFmode or TCmode.
3285 (FLOAT128_IBM_P): Likewise.
3287 2018-01-10 Martin Sebor <msebor@redhat.com>
3289 PR tree-optimization/83671
3290 * builtins.c (c_strlen): Unconditionally return zero for the empty
3292 Use -Warray-bounds for warnings.
3293 * gimple-fold.c (get_range_strlen): Handle non-constant lengths
3294 for non-constant array indices with COMPONENT_REF, arrays of
3295 arrays, and pointers to arrays.
3296 (gimple_fold_builtin_strlen): Determine and set length range for
3297 non-constant character arrays.
3299 2018-01-10 Aldy Hernandez <aldyh@redhat.com>
3302 * tree-ssa-uninit.c (convert_control_dep_chain_into_preds): Skip
3305 2018-01-10 Eric Botcazou <ebotcazou@adacore.com>
3307 * dwarf2out.c (dwarf2out_var_location): Do not pass NULL to fprintf.
3309 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3312 * config/rs6000/rs6000.c (print_operand) <'y'>: Use
3313 VECTOR_MEM_ALTIVEC_OR_VSX_P.
3314 * config/rs6000/vsx.md (*vsx_le_perm_load_<mode> for VSX_D): Use
3315 indexed_or_indirect_operand predicate.
3316 (*vsx_le_perm_load_<mode> for VSX_W): Likewise.
3317 (*vsx_le_perm_load_v8hi): Likewise.
3318 (*vsx_le_perm_load_v16qi): Likewise.
3319 (*vsx_le_perm_store_<mode> for VSX_D): Likewise.
3320 (*vsx_le_perm_store_<mode> for VSX_W): Likewise.
3321 (*vsx_le_perm_store_v8hi): Likewise.
3322 (*vsx_le_perm_store_v16qi): Likewise.
3323 (eight unnamed splitters): Likewise.
3325 2018-01-10 Peter Bergner <bergner@vnet.ibm.com>
3327 * config/rs6000/x86intrin.h: Change #warning to #error. Update message.
3328 * config/rs6000/emmintrin.h: Likewise.
3329 * config/rs6000/mmintrin.h: Likewise.
3330 * config/rs6000/xmmintrin.h: Likewise.
3332 2018-01-10 David Malcolm <dmalcolm@redhat.com>
3335 * tree-core.h: Document EXPR_LOCATION_WRAPPER_P's usage of
3337 * tree.c (tree_nop_conversion): Return true for location wrapper
3339 (maybe_wrap_with_location): New function.
3340 (selftest::check_strip_nops): New function.
3341 (selftest::test_location_wrappers): New function.
3342 (selftest::tree_c_tests): Call it.
3343 * tree.h (STRIP_ANY_LOCATION_WRAPPER): New macro.
3344 (maybe_wrap_with_location): New decl.
3345 (EXPR_LOCATION_WRAPPER_P): New macro.
3346 (location_wrapper_p): New inline function.
3347 (tree_strip_any_location_wrapper): New inline function.
3349 2018-01-10 H.J. Lu <hongjiu.lu@intel.com>
3352 * config/i386/i386.c (ix86_compute_frame_layout): Always adjust
3353 stack_realign_offset for the largest alignment of stack slot
3355 (ix86_find_max_used_stack_alignment): New function.
3356 (ix86_finalize_stack_frame_flags): Use it. Set
3357 max_used_stack_alignment if we don't realign stack.
3358 * config/i386/i386.h (machine_function): Add
3359 max_used_stack_alignment.
3361 2018-01-10 Christophe Lyon <christophe.lyon@linaro.org>
3363 * config/arm/arm.opt (-mbranch-cost): New option.
3364 * config/arm/arm.h (BRANCH_COST): Take arm_branch_cost into
3367 2018-01-10 Segher Boessenkool <segher@kernel.crashing.org>
3370 * config/rs6000/rs6000.md (load_toc_v4_PIC_2, load_toc_v4_PIC_3b,
3371 load_toc_v4_PIC_3c): Wrap const term in CONST RTL.
3373 2018-01-10 Richard Biener <rguenther@suse.de>
3376 * dwarf2out.c (gen_subprogram_die): Hoist old_die && declaration
3377 early out so it also covers the case where we have a non-NULL
3380 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3382 PR tree-optimization/83753
3383 * tree-vect-stmts.c (get_group_load_store_type): Use VMAT_CONTIGUOUS
3384 for non-strided grouped accesses if the number of elements is 1.
3386 2018-01-10 Jan Hubicka <hubicka@ucw.cz>
3389 * i386.c (ix86_vectorize_builtin_gather): Check TARGET_USE_GATHER.
3390 * i386.h (TARGET_USE_GATHER): Define.
3391 * x86-tune.def (X86_TUNE_USE_GATHER): New.
3393 2018-01-10 Martin Liska <mliska@suse.cz>
3396 * basic-block.h (CLEANUP_NO_PARTITIONING): New define.
3397 * bb-reorder.c (pass_reorder_blocks::execute): Do not clean up
3399 * cfgcleanup.c (try_optimize_cfg): Fix up partitioning if
3400 CLEANUP_NO_PARTITIONING is not set.
3402 2018-01-10 Richard Sandiford <richard.sandiford@linaro.org>
3404 * doc/rtl.texi: Remove documentation of (const ...) wrappers
3405 for vectors, as a partial revert of r254296.
3406 * rtl.h (const_vec_p): Delete.
3407 (const_vec_duplicate_p): Don't test for vector CONSTs.
3408 (unwrap_const_vec_duplicate, const_vec_series_p): Likewise.
3409 * expmed.c (make_tree): Likewise.
3412 * common.md (E, F): Use CONSTANT_P instead of checking for
3414 * emit-rtl.c (gen_lowpart_common): Use const_vec_p instead of
3415 checking for CONST_VECTOR.
3417 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3420 * predict.c (force_edge_cold): Handle in more sane way edges
3423 2018-01-09 Carl Love <cel@us.ibm.com>
3425 * config/rs6002/altivec.md (p8_vmrgow): Add support for V2DI, V2DF,
3427 (p8_vmrgew): Add support for V2DI, V2DF, V4SF types.
3428 * config/rs6000/rs6000-builtin.def: Add definitions for FLOAT2_V2DF,
3429 VMRGEW_V2DI, VMRGEW_V2DF, VMRGEW_V4SF, VMRGOW_V4SI, VMRGOW_V4SF,
3430 VMRGOW_V2DI, VMRGOW_V2DF. Remove definition for VMRGOW.
3431 * config/rs6000/rs6000-c.c (VSX_BUILTIN_VEC_FLOAT2,
3432 P8V_BUILTIN_VEC_VMRGEW, P8V_BUILTIN_VEC_VMRGOW): Add definitions.
3433 * config/rs6000/rs6000-protos.h: Add extern defition for
3434 rs6000_generate_float2_double_code.
3435 * config/rs6000/rs6000.c (rs6000_generate_float2_double_code): Add
3437 * config/rs6000/vsx.md (vsx_xvcdpsp): Add define_insn.
3438 (float2_v2df): Add define_expand.
3440 2018-01-09 Uros Bizjak <ubizjak@gmail.com>
3443 * combine.c (force_int_to_mode) <case ASHIFT>: Use mode instead of
3444 op_mode in the force_to_mode call.
3446 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3448 * config/aarch64/aarch64.c (aarch64_evpc_trn): Use d.perm.series_p
3449 instead of checking each element individually.
3450 (aarch64_evpc_uzp): Likewise.
3451 (aarch64_evpc_zip): Likewise.
3452 (aarch64_evpc_ext): Likewise.
3453 (aarch64_evpc_rev): Likewise.
3454 (aarch64_evpc_dup): Test the encoding for a single duplicated element,
3455 instead of checking each element individually. Return true without
3457 (aarch64_vectorize_vec_perm_const): Use all_from_input_p to test
3458 whether all selected elements come from the same input, instead of
3459 checking each element individually. Remove calls to gen_rtx_REG,
3460 start_sequence and end_sequence and instead assert that no rtl is
3463 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3465 * config/aarch64/aarch64.c (aarch64_legitimate_constant_p): Fix
3466 order of HIGH and CONST checks.
3468 2018-01-09 Richard Sandiford <richard.sandiford@linaro.org>
3470 * tree-vect-stmts.c (permute_vec_elements): Create a fresh variable
3471 if the destination isn't an SSA_NAME.
3473 2018-01-09 Richard Biener <rguenther@suse.de>
3475 PR tree-optimization/83668
3476 * graphite.c (canonicalize_loop_closed_ssa): Add edge argument,
3478 (canonicalize_loop_form): ... here, renamed from ...
3479 (canonicalize_loop_closed_ssa_form): ... this and amended to
3480 swap successor edges for loop exit blocks to make us use
3481 the RPO order we need for initial schedule generation.
3483 2018-01-09 Joseph Myers <joseph@codesourcery.com>
3485 PR tree-optimization/64811
3486 * match.pd: When optimizing comparisons with Inf, avoid
3487 introducing or losing exceptions from comparisons with NaN.
3489 2018-01-09 Martin Liska <mliska@suse.cz>
3492 * asan.c (shadow_mem_size): Add gcc_assert.
3494 2018-01-09 Georg-Johann Lay <avr@gjlay.de>
3496 Don't save registers in main().
3499 * doc/invoke.texi (AVR Options) [-mmain-is-OS_task]: Document it.
3500 * config/avr/avr.opt (-mmain-is-OS_task): New target option.
3501 * config/avr/avr.c (avr_set_current_function): Don't error if
3502 naked, OS_task or OS_main are specified at the same time.
3503 (avr_function_ok_for_sibcall): Don't disable sibcalls for OS_task,
3505 (avr_insert_attributes) [-mmain-is-OS_task] <main>: Add OS_task
3507 * common/config/avr/avr-common.c (avr_option_optimization_table):
3508 Switch on -mmain-is-OS_task for optimizing compilations.
3510 2018-01-09 Richard Biener <rguenther@suse.de>
3512 PR tree-optimization/83572
3513 * graphite.c: Include cfganal.h.
3514 (graphite_transform_loops): Connect infinite loops to exit
3515 and remove fake edges at the end.
3517 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3519 * ipa-inline.c (edge_badness): Revert accidental checkin.
3521 2018-01-09 Jan Hubicka <hubicka@ucw.cz>
3524 * ipa-comdats.c (set_comdat_group): Only set comdat group of real
3525 symbols; not inline clones.
3527 2018-01-09 Jakub Jelinek <jakub@redhat.com>
3530 * modulo-sched.c (schedule_reg_moves): Punt if we'd need to move
3531 hard registers. Formatting fixes.
3533 PR preprocessor/83722
3534 * gcc.c (try_generate_repro): Pass
3535 &temp_stderr_files[RETRY_ICE_ATTEMPTS - 1] rather than
3536 &temp_stdout_files[RETRY_ICE_ATTEMPTS - 1] as last argument to
3539 2018-01-08 Monk Chiang <sh.chiang04@gmail.com>
3540 Kito Cheng <kito.cheng@gmail.com>
3542 * config/riscv/riscv.c (machine_function::is_leaf): Remove field.
3543 (riscv_leaf_function_p): Delete.
3544 (riscv_function_ok_for_sibcall): Return false when TARGET_SAVE_RESTORE.
3546 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3548 * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
3550 (do_ifelse): New function.
3551 (do_isel): New function.
3552 (do_sub3): New function.
3553 (do_add3): New function.
3554 (do_load_mask_compare): New function.
3555 (do_overlap_load_compare): New function.
3556 (expand_compare_loop): New function.
3557 (expand_block_compare): Call expand_compare_loop() when appropriate.
3558 * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
3560 (-mblock-compare-inline-loop-limit): New option.
3562 2018-01-08 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
3565 * config/rs6000/altivec.md (*altivec_vpermr_<mode>_internal):
3566 Reverse order of second and third operands in first alternative.
3567 * config/rs6000/rs6000.c (rs6000_expand_vector_set): Reverse order
3568 of first and second elements in UNSPEC_VPERMR vector.
3569 (altivec_expand_vec_perm_le): Likewise.
3571 2017-01-08 Jeff Law <law@redhat.com>
3573 PR rtl-optimizatin/81308
3574 * tree-switch-conversion.c (cfg_altered): New file scoped static.
3575 (process_switch): If group_case_labels makes a change, then set
3577 (pass_convert_switch::execute): If a switch is converted, then
3578 set cfg_altered. Return TODO_cfg_cleanup if cfg_altered is true.
3580 PR rtl-optimization/81308
3581 * recog.c (split_all_insns): Conditionally cleanup the CFG after
3584 2018-01-08 Vidya Praveen <vidyapraveen@arm.com>
3586 PR target/83663 - Revert r255946
3587 * config/aarch64/aarch64.c (aarch64_expand_vector_init): Modify code
3588 generation for cases where splatting a value is not useful.
3589 * simplify-rtx.c (simplify_ternary_operation): Simplify vec_merge
3590 across a vec_duplicate and a paradoxical subreg forming a vector
3591 mode to a vec_concat.
3593 2018-01-08 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
3595 * config/arm/t-aprofile (MULTILIB_MATCHES): Add mapping rules for
3596 -march=armv8.3-a variants.
3597 * config/arm/t-multilib: Likewise.
3598 * config/arm/t-arm-elf: Likewise. Handle dotprod extension.
3600 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
3602 * config/rs6000/rs6000.md (cceq_ior_compare): Remove * so I can use it
3604 (cceq_ior_compare_complement): Give it a name so I can use it, and
3605 change boolean_or_operator predicate to boolean_operator so it can
3606 be used to generate a crand.
3607 (eqne): New code iterator.
3608 (bd/bd_neg): New code_attrs.
3609 (<bd>_<mode>): New name for ctr<mode>_internal[12] now combined into
3610 a single define_insn.
3611 (<bd>tf_<mode>): A new insn pattern for the conditional form branch
3612 decrement (bdnzt/bdnzf/bdzt/bdzf).
3613 * config/rs6000/rs6000.c (rs6000_legitimate_combined_insn): Updated
3614 with the new names of the branch decrement patterns, and added the
3615 names of the branch decrement conditional patterns.
3617 2018-01-08 Richard Biener <rguenther@suse.de>
3619 PR tree-optimization/83563
3620 * graphite.c (canonicalize_loop_closed_ssa_form): Reset the SCEV
3623 2018-01-08 Richard Biener <rguenther@suse.de>
3626 * convert.c (do_narrow): Properly guard TYPE_OVERFLOW_WRAPS checks.
3628 2018-01-08 Richard Biener <rguenther@suse.de>
3630 PR tree-optimization/83685
3631 * tree-ssa-pre.c (create_expression_by_pieces): Do not insert
3632 references to abnormals.
3634 2018-01-08 Richard Biener <rguenther@suse.de>
3637 * dwarf2out.c (output_indirect_strings): Handle empty
3638 skeleton_debug_str_hash.
3639 (dwarf2out_early_finish): Index strings for -gsplit-dwarf.
3641 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3643 * config/arc/arc.c (TARGET_TRAMPOLINE_ADJUST_ADDRESS): Delete.
3644 (emit_store_direct): Likewise.
3645 (arc_trampoline_adjust_address): Likewise.
3646 (arc_asm_trampoline_template): New function.
3647 (arc_initialize_trampoline): Use asm_trampoline_template.
3648 (TARGET_ASM_TRAMPOLINE_TEMPLATE): Define.
3649 * config/arc/arc.h (TRAMPOLINE_SIZE): Adjust to 16.
3650 * config/arc/arc.md (flush_icache): Delete pattern.
3652 2018-01-08 Claudiu Zissulescu <claziss@synopsys.com>
3654 * config/arc/arc-c.def (__ARC_UNALIGNED__): New define.
3655 * config/arc/arc.h (STRICT_ALIGNMENT): Control this macro using
3658 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3661 * config/epiphany/epiphany.h (make_pass_mode_switch_use): Guard
3662 by not USED_FOR_TARGET.
3663 (make_pass_resolve_sw_modes): Likewise.
3665 2018-01-08 Sebastian Huber <sebastian.huber@embedded-brains.de>
3667 * config/nios2/nios2.h (nios2_section_threshold): Guard by not
3670 2018-01-08 Richard Biener <rguenther@suse.de>
3673 * tree-data-ref.c (split_constant_offset): Remove STRIP_NOPS.
3675 2018-01-08 Richard Biener <rguenther@suse.de>
3678 * match.pd ((t * 2) / 2) -> t): Add missing :c.
3680 2018-01-06 Aldy Hernandez <aldyh@redhat.com>
3683 * tree-ssa-uninit.c (compute_control_dep_chain): Do not bail on
3684 basic blocks with a small number of successors.
3685 (convert_control_dep_chain_into_preds): Improve handling of
3687 (dump_predicates): Split apart into...
3688 (dump_pred_chain): ...here...
3689 (dump_pred_info): ...and here.
3690 (can_one_predicate_be_invalidated_p): Add debugging printfs.
3691 (can_chain_union_be_invalidated_p): Improve check for invalidation
3693 (uninit_uses_cannot_happen): Avoid unnecessary if
3694 convert_control_dep_chain_into_preds yielded nothing.
3696 2018-01-06 Martin Sebor <msebor@redhat.com>
3698 PR tree-optimization/83640
3699 * gimple-ssa-warn-restrict.c (builtin_access::builtin_access): Avoid
3700 subtracting negative offset from size.
3701 (builtin_access::overlap): Adjust offset bounds of the access to fall
3702 within the size of the object if possible.
3704 2018-01-06 Richard Sandiford <richard.sandiford@linaro.org>
3706 PR rtl-optimization/83699
3707 * expmed.c (extract_bit_field_1): Restrict the vector usage of
3708 extract_bit_field_as_subreg to cases in which the extracted
3709 value is also a vector.
3711 * lra-constraints.c (process_alt_operands): Test for the equivalence
3712 substitutions when detecting a possible reload cycle.
3714 2018-01-06 Jakub Jelinek <jakub@redhat.com>
3717 * toplev.c (process_options): Don't enable debug_nonbind_markers_p
3718 by default if flag_selective_schedling{,2}. Formatting fixes.
3720 PR rtl-optimization/83682
3721 * rtl.h (const_vec_duplicate_p): Only return true for VEC_DUPLICATE
3722 if it has non-VECTOR_MODE element mode.
3723 (vec_duplicate_p): Likewise.
3726 * cfgexpand.c (expand_debug_expr): Punt if mode1 is VOIDmode
3727 and bitsize might be greater than MAX_BITSIZE_MODE_ANY_INT.
3729 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3732 * config/i386/i386-builtin.def
3733 (__builtin_ia32_vgf2p8affineinvqb_v64qi,
3734 __builtin_ia32_vgf2p8affineqb_v64qi, __builtin_ia32_vgf2p8mulb_v64qi):
3735 Require also OPTION_MASK_ISA_AVX512F in addition to
3736 OPTION_MASK_ISA_GFNI.
3737 (__builtin_ia32_vgf2p8affineinvqb_v16qi_mask,
3738 __builtin_ia32_vgf2p8affineqb_v16qi_mask): Require
3739 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_SSE in addition
3740 to OPTION_MASK_ISA_GFNI.
3741 (__builtin_ia32_vgf2p8mulb_v32qi_mask): Require
3742 OPTION_MASK_ISA_AVX512VL in addition to OPTION_MASK_ISA_GFNI and
3743 OPTION_MASK_ISA_AVX512BW.
3744 (__builtin_ia32_vgf2p8mulb_v16qi_mask): Require
3745 OPTION_MASK_ISA_AVX512VL instead of OPTION_MASK_ISA_AVX512BW in
3746 addition to OPTION_MASK_ISA_GFNI.
3747 (__builtin_ia32_vgf2p8affineinvqb_v16qi,
3748 __builtin_ia32_vgf2p8affineqb_v16qi, __builtin_ia32_vgf2p8mulb_v16qi):
3749 Require OPTION_MASK_ISA_SSE2 instead of OPTION_MASK_ISA_SSE in addition
3750 to OPTION_MASK_ISA_GFNI.
3751 * config/i386/i386.c (def_builtin): Change to builtin isa/isa2 being
3752 a requirement for all ISAs rather than any of them with a few
3754 (ix86_add_new_builtins): Clear OPTION_MASK_ISA_64BIT from isa before
3756 (ix86_expand_builtin): Require all ISAs from builtin's isa and isa2
3757 bitmasks to be enabled with 3 exceptions, instead of requiring any
3758 enabled ISA with lots of exceptions.
3759 * config/i386/sse.md (vgf2p8affineinvqb_<mode><mask_name>,
3760 vgf2p8affineqb_<mode><mask_name>, vgf2p8mulb_<mode><mask_name>):
3761 Change avx512bw in isa attribute to avx512f.
3762 * config/i386/sgxintrin.h: Add license boilerplate.
3763 * config/i386/vaesintrin.h: Likewise. Fix macro spelling __AVX512F
3764 to __AVX512F__ and __AVX512VL to __AVX512VL__.
3765 (_mm256_aesdec_epi128, _mm256_aesdeclast_epi128, _mm256_aesenc_epi128,
3766 _mm256_aesenclast_epi128): Enable temporarily avx if __AVX__ is not
3768 * config/i386/gfniintrin.h (_mm_gf2p8mul_epi8,
3769 _mm_gf2p8affineinv_epi64_epi8, _mm_gf2p8affine_epi64_epi8): Enable
3770 temporarily sse2 rather than sse if not enabled already.
3773 * config/i386/sse.md (VI248_VLBW): Rename to ...
3774 (VI248_AVX512VL): ... this. Don't guard V32HI with TARGET_AVX512BW.
3775 (vpshrd_<mode><mask_name>, vpshld_<mode><mask_name>,
3776 vpshrdv_<mode>, vpshrdv_<mode>_mask, vpshrdv_<mode>_maskz,
3777 vpshrdv_<mode>_maskz_1, vpshldv_<mode>, vpshldv_<mode>_mask,
3778 vpshldv_<mode>_maskz, vpshldv_<mode>_maskz_1): Use VI248_AVX512VL
3779 mode iterator instead of VI248_VLBW.
3781 2018-01-05 Jan Hubicka <hubicka@ucw.cz>
3783 * ipa-fnsummary.c (record_modified_bb_info): Add OP.
3784 (record_modified): Skip clobbers; add debug output.
3785 (param_change_prob): Use sreal frequencies.
3787 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3789 * tree-vect-data-refs.c (vect_compute_data_ref_alignment): Don't
3790 punt for user-aligned variables.
3792 2018-01-05 Richard Sandiford <richard.sandiford@linaro.org>
3794 * tree-chrec.c (chrec_contains_symbols): Return true for
3797 2018-01-05 Sudakshina Das <sudi.das@arm.com>
3800 * simplify-rtx.c (simplify_relational_operation_1): Add simplifications
3801 of (x|y) == x for BICS pattern.
3803 2018-01-05 Jakub Jelinek <jakub@redhat.com>
3805 PR tree-optimization/83605
3806 * gimple-ssa-strength-reduction.c: Include tree-eh.h.
3807 (find_candidates_dom_walker::before_dom_children): Ignore stmts that
3810 2018-01-05 Sebastian Huber <sebastian.huber@embedded-brains.de>
3812 * config.gcc (epiphany-*-elf*): Add (epiphany-*-rtems*) configuration.
3813 * config/epiphany/rtems.h: New file.
3815 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3816 Uros Bizjak <ubizjak@gmail.com>
3819 * config/i386/i386.md (*<rotate_insn>hi3_1 splitter): Use
3820 QIreg_operand instead of register_operand predicate.
3821 * config/i386/i386.c (ix86_rop_should_change_byte_p,
3822 set_rop_modrm_reg_bits, ix86_mitigate_rop): Use -mmitigate-rop in
3823 comments instead of -fmitigate[-_]rop.
3825 2018-01-04 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
3828 * cgraphunit.c (symbol_table::compile): Switch to text_section
3829 before calling assembly_start debug hook.
3830 * run-rtl-passes.c (run_rtl_passes): Likewise.
3833 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3835 * tree-vrp.c (extract_range_from_binary_expr_1): Check
3836 range_int_cst_p rather than !symbolic_range_p before calling
3837 extract_range_from_multiplicative_op_1.
3839 2017-01-04 Jeff Law <law@redhat.com>
3841 * tree-ssa-math-opts.c (execute_cse_reciprocals_1): Remove
3842 redundant test in assertion.
3844 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3846 * doc/rtl.texi: Document machine_mode wrapper classes.
3848 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3850 * fold-const.c (fold_ternary_loc): Check tree_fits_uhwi_p before
3853 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3855 * tree-ssa-forwprop.c (is_combined_permutation_identity): Allow
3856 the VEC_PERM_EXPR fold to fail.
3858 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3861 * bb-reorder.c (insert_section_boundary_note): Set has_bb_partition
3862 to switched_sections.
3864 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3867 * config/arm/arm.c (arm_vectorize_vec_perm_const): Fix inverted
3870 2018-01-04 Peter Bergner <bergner@vnet.ibm.com>
3873 * config/rs6000/rs6000.c (rs6000_discover_homogeneous_aggregate): Do not
3874 allow arguments in FP registers if TARGET_HARD_FLOAT is false.
3876 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3879 * cfgexpand.c (expand_debug_expr) <case BIT_FIELD_REF>: Punt if mode
3880 is BLKmode and bitpos not zero or mode change is needed.
3882 2018-01-04 Richard Sandiford <richard.sandiford@linaro.org>
3885 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): Require
3888 2018-01-04 Uros Bizjak <ubizjak@gmail.com>
3891 * config/alpha/alpha.md (*sadd<modesuffix>): Use ASHIFT
3892 instead of MULT rtx. Update all corresponding splitters.
3894 (*ssub<modesuffix>): Ditto.
3896 (*cmp_sadd_di): Update split patterns.
3897 (*cmp_sadd_si): Ditto.
3898 (*cmp_sadd_sidi): Ditto.
3899 (*cmp_ssub_di): Ditto.
3900 (*cmp_ssub_si): Ditto.
3901 (*cmp_ssub_sidi): Ditto.
3902 * config/alpha/predicates.md (const23_operand): New predicate.
3903 * config/alpha/alpha.c (alpha_rtx_costs) [PLUS, MINUS]:
3904 Look for ASHIFT, not MULT inner operand.
3905 (alpha_split_conditional_move): Update for *sadd<modesuffix> change.
3907 2018-01-04 Martin Liska <mliska@suse.cz>
3909 PR gcov-profile/83669
3910 * gcov.c (output_intermediate_file): Add version to intermediate
3912 * doc/gcov.texi: Document new field 'version' in intermediate
3913 file format. Fix location of '-k' option of gcov command.
3915 2018-01-04 Martin Liska <mliska@suse.cz>
3918 * ipa-icf.c (sem_function::merge): Do not cross comdat boundary.
3920 2018-01-04 Jakub Jelinek <jakub@redhat.com>
3922 * gimple-ssa-sprintf.c (parse_directive): Cast second dir.len to uhwi.
3924 2018-01-03 Martin Sebor <msebor@redhat.com>
3926 PR tree-optimization/83655
3927 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call): Avoid
3928 checking calls with invalid arguments.
3930 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3932 * tree-vect-stmts.c (vect_get_store_rhs): New function.
3933 (vectorizable_mask_load_store): Delete.
3934 (vectorizable_call): Return false for masked loads and stores.
3935 (vectorizable_store): Handle IFN_MASK_STORE. Use vect_get_store_rhs
3936 instead of gimple_assign_rhs1.
3937 (vectorizable_load): Handle IFN_MASK_LOAD.
3938 (vect_transform_stmt): Don't set is_store for call_vec_info_type.
3940 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3942 * tree-vect-stmts.c (vect_build_gather_load_calls): New function,
3944 (vectorizable_mask_load_store): ...here.
3945 (vectorizable_load): ...and here.
3947 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3949 * tree-vect-stmts.c (vect_build_all_ones_mask)
3950 (vect_build_zero_merge_argument): New functions, split out from...
3951 (vectorizable_load): ...here.
3953 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3955 * tree-vect-stmts.c (vect_check_store_rhs): New function,
3957 (vectorizable_mask_load_store): ...here.
3958 (vectorizable_store): ...and here.
3960 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3962 * tree-vect-stmts.c (vect_check_load_store_mask): New function,
3964 (vectorizable_mask_load_store): ...here.
3966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3968 * tree-vectorizer.h (vec_load_store_type): Moved from tree-vec-stmts.c
3969 (vect_model_store_cost): Take a vec_load_store_type instead of a
3971 * tree-vect-stmts.c (vec_load_store_type): Move to tree-vectorizer.h.
3972 (vect_model_store_cost): Take a vec_load_store_type instead of a
3974 (vectorizable_mask_load_store): Update accordingly.
3975 (vectorizable_store): Likewise.
3976 * tree-vect-slp.c (vect_analyze_slp_cost_1): Update accordingly.
3978 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3980 * tree-vect-loop.c (vect_transform_loop): Stub out scalar
3981 IFN_MASK_LOAD calls here rather than...
3982 * tree-vect-stmts.c (vectorizable_mask_load_store): ...here.
3984 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3985 Alan Hayward <alan.hayward@arm.com>
3986 David Sherwood <david.sherwood@arm.com>
3988 * expmed.c (extract_bit_field_1): For vector extracts,
3989 fall back to extract_bit_field_as_subreg if vec_extract
3992 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
3993 Alan Hayward <alan.hayward@arm.com>
3994 David Sherwood <david.sherwood@arm.com>
3996 * lra-spills.c (pseudo_reg_slot_compare): Sort slots by whether
3997 they are variable or constant sized.
3998 (assign_stack_slot_num_and_sort_pseudos): Don't reuse variable-sized
3999 slots for constant-sized data.
4001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4002 Alan Hayward <alan.hayward@arm.com>
4003 David Sherwood <david.sherwood@arm.com>
4005 * tree-vect-patterns.c (vect_recog_mask_conversion_pattern): When
4006 handling COND_EXPRs with boolean comparisons, try to find a better
4007 basis for the mask type than the boolean itself.
4009 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4011 * doc/rtl.texi (MAX_BITSIZE_MODE_ANY_MODE): Describe how the default
4012 is calculated and how it can be overridden.
4013 * genmodes.c (max_bitsize_mode_any_mode): New variable.
4014 (create_modes): Initialize it from MAX_BITSIZE_MODE_ANY_MODE,
4016 (emit_max_int): Use it to set the output MAX_BITSIZE_MODE_ANY_MODE,
4019 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4020 Alan Hayward <alan.hayward@arm.com>
4021 David Sherwood <david.sherwood@arm.com>
4023 * config/aarch64/aarch64-protos.h (aarch64_output_simd_mov_immediate):
4024 Remove the mode argument.
4025 (aarch64_simd_valid_immediate): Remove the mode and inverse
4027 * config/aarch64/iterators.md (bitsize): New iterator.
4028 * config/aarch64/aarch64-simd.md (*aarch64_simd_mov<mode>, and<mode>3)
4029 (ior<mode>3): Update calls to aarch64_output_simd_mov_immediate.
4030 * config/aarch64/constraints.md (Do, Db, Dn): Update calls to
4031 aarch64_simd_valid_immediate.
4032 * config/aarch64/predicates.md (aarch64_reg_or_orr_imm): Likewise.
4033 (aarch64_reg_or_bic_imm): Likewise.
4034 * config/aarch64/aarch64.c (simd_immediate_info): Replace mvn
4035 with an insn_type enum and msl with a modifier_type enum.
4036 Replace element_width with a scalar_mode. Change the shift
4037 to unsigned int. Add constructors for scalar_float_mode and
4038 scalar_int_mode elements.
4039 (aarch64_vect_float_const_representable_p): Delete.
4040 (aarch64_can_const_movi_rtx_p)
4041 (aarch64_simd_scalar_immediate_valid_for_move)
4042 (aarch64_simd_make_constant): Update call to
4043 aarch64_simd_valid_immediate.
4044 (aarch64_advsimd_valid_immediate_hs): New function.
4045 (aarch64_advsimd_valid_immediate): Likewise.
4046 (aarch64_simd_valid_immediate): Remove mode and inverse
4047 arguments. Rewrite to use the above. Use const_vec_duplicate_p
4048 to detect duplicated constants and use aarch64_float_const_zero_rtx_p
4049 and aarch64_float_const_representable_p on the result.
4050 (aarch64_output_simd_mov_immediate): Remove mode argument.
4051 Update call to aarch64_simd_valid_immediate and use of
4052 simd_immediate_info.
4053 (aarch64_output_scalar_simd_mov_immediate): Update call
4056 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4057 Alan Hayward <alan.hayward@arm.com>
4058 David Sherwood <david.sherwood@arm.com>
4060 * machmode.h (mode_precision): Prefix with CONST_MODE_PRECISION.
4061 (mode_nunits): Likewise CONST_MODE_NUNITS.
4062 * machmode.def (ADJUST_NUNITS): Document.
4063 * genmodes.c (mode_data::need_nunits_adj): New field.
4064 (blank_mode): Update accordingly.
4065 (adj_nunits): New variable.
4066 (print_maybe_const_decl): Replace CATEGORY with a NEEDS_ADJ
4068 (emit_mode_size_inline): Set need_bytesize_adj for all modes
4069 listed in adj_nunits.
4070 (emit_mode_nunits_inline): Set need_nunits_adj for all modes
4071 listed in adj_nunits. Don't emit case statements for such modes.
4072 (emit_insn_modes_h): Emit definitions of CONST_MODE_NUNITS
4073 and CONST_MODE_PRECISION. Make CONST_MODE_SIZE expand to
4074 nothing if adj_nunits is nonnull.
4075 (emit_mode_precision, emit_mode_nunits): Use print_maybe_const_decl.
4076 (emit_mode_unit_size, emit_mode_base_align, emit_mode_ibit)
4077 (emit_mode_fbit): Update use of print_maybe_const_decl.
4078 (emit_move_size): Likewise. Treat the array as non-const
4080 (emit_mode_adjustments): Handle adj_nunits.
4082 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4084 * machmode.def (VECTOR_MODES_WITH_PREFIX): Document.
4085 * genmodes.c (VECTOR_MODES_WITH_PREFIX): New macro.
4086 (VECTOR_MODES): Use it.
4087 (make_vector_modes): Take the prefix as an argument.
4089 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4090 Alan Hayward <alan.hayward@arm.com>
4091 David Sherwood <david.sherwood@arm.com>
4093 * mode-classes.def (MODE_VECTOR_BOOL): New mode class.
4094 * machmode.h (INTEGRAL_MODE_P, VECTOR_MODE_P): Return true
4095 for MODE_VECTOR_BOOL.
4096 * machmode.def (VECTOR_BOOL_MODE): Document.
4097 * genmodes.c (VECTOR_BOOL_MODE): New macro.
4098 (make_vector_bool_mode): New function.
4099 (complete_mode, emit_mode_wider, emit_mode_adjustments): Handle
4101 * lto-streamer-in.c (lto_input_mode_table): Likewise.
4102 * rtx-vector-builder.c (rtx_vector_builder::find_cached_value):
4104 * stor-layout.c (int_mode_for_mode): Likewise.
4105 * tree.c (build_vector_type_for_mode): Likewise.
4106 * varasm.c (output_constant_pool_2): Likewise.
4107 * emit-rtl.c (init_emit_once): Make sure that CONST1_RTX (BImode) and
4108 CONSTM1_RTX (BImode) are the same thing. Initialize const_tiny_rtx
4109 for MODE_VECTOR_BOOL.
4110 * expr.c (expand_expr_real_1): Use VECTOR_MODE_P instead of a list
4111 of mode class checks.
4112 * tree-vect-generic.c (expand_vector_operation): Use VECTOR_MODE_P
4113 instead of a list of mode class checks.
4114 (expand_vector_scalar_condition): Likewise.
4115 (type_for_widest_vector_mode): Handle BImode as an inner mode.
4117 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4118 Alan Hayward <alan.hayward@arm.com>
4119 David Sherwood <david.sherwood@arm.com>
4121 * machmode.h (mode_size): Change from unsigned short to
4123 (mode_to_bytes): Return a poly_uint16 rather than an unsigned short.
4124 (GET_MODE_SIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4125 or if measurement_type is not polynomial.
4126 (fixed_size_mode::includes_p): Check for constant-sized modes.
4127 * genmodes.c (emit_mode_size_inline): Make mode_size_inline
4128 return a poly_uint16 rather than an unsigned short.
4129 (emit_mode_size): Change the type of mode_size from unsigned short
4130 to poly_uint16_pod. Use ZERO_COEFFS for the initializer.
4131 (emit_mode_adjustments): Cope with polynomial vector sizes.
4132 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4134 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4136 * auto-inc-dec.c (try_merge): Treat GET_MODE_SIZE as polynomial.
4137 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Likewise.
4138 * caller-save.c (setup_save_areas): Likewise.
4139 (replace_reg_with_saved_mem): Likewise.
4140 * calls.c (emit_library_call_value_1): Likewise.
4141 * combine-stack-adj.c (combine_stack_adjustments_for_block): Likewise.
4142 * combine.c (simplify_set, make_extraction, simplify_shift_const_1)
4143 (gen_lowpart_for_combine): Likewise.
4144 * convert.c (convert_to_integer_1): Likewise.
4145 * cse.c (equiv_constant, cse_insn): Likewise.
4146 * cselib.c (autoinc_split, cselib_hash_rtx): Likewise.
4147 (cselib_subst_to_values): Likewise.
4148 * dce.c (word_dce_process_block): Likewise.
4149 * df-problems.c (df_word_lr_mark_ref): Likewise.
4150 * dwarf2cfi.c (init_one_dwarf_reg_size): Likewise.
4151 * dwarf2out.c (multiple_reg_loc_descriptor, mem_loc_descriptor)
4152 (concat_loc_descriptor, concatn_loc_descriptor, loc_descriptor)
4153 (rtl_for_decl_location): Likewise.
4154 * emit-rtl.c (gen_highpart, widen_memory_access): Likewise.
4155 * expmed.c (extract_bit_field_1, extract_integral_bit_field): Likewise.
4156 * expr.c (emit_group_load_1, clear_storage_hints): Likewise.
4157 (emit_move_complex, emit_move_multi_word, emit_push_insn): Likewise.
4158 (expand_expr_real_1): Likewise.
4159 * function.c (assign_parm_setup_block_p, assign_parm_setup_block)
4160 (pad_below): Likewise.
4161 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4162 * gimple-ssa-store-merging.c (rhs_valid_for_store_merging_p): Likewise.
4163 * ira.c (get_subreg_tracking_sizes): Likewise.
4164 * ira-build.c (ira_create_allocno_objects): Likewise.
4165 * ira-color.c (coalesced_pseudo_reg_slot_compare): Likewise.
4166 (ira_sort_regnos_for_alter_reg): Likewise.
4167 * ira-costs.c (record_operand_costs): Likewise.
4168 * lower-subreg.c (interesting_mode_p, simplify_gen_subreg_concatn)
4169 (resolve_simple_move): Likewise.
4170 * lra-constraints.c (get_reload_reg, operands_match_p): Likewise.
4171 (process_addr_reg, simplify_operand_subreg, curr_insn_transform)
4172 (lra_constraints): Likewise.
4173 (CONST_POOL_OK_P): Reject variable-sized modes.
4174 * lra-spills.c (slot, assign_mem_slot, pseudo_reg_slot_compare)
4175 (add_pseudo_to_slot, lra_spill): Likewise.
4176 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4177 * optabs-query.c (get_best_extraction_insn): Likewise.
4178 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4179 * optabs.c (expand_vec_perm_var, expand_vec_cond_expr): Likewise.
4180 (expand_mult_highpart, valid_multiword_target_p): Likewise.
4181 * recog.c (offsettable_address_addr_space_p): Likewise.
4182 * regcprop.c (maybe_mode_change): Likewise.
4183 * reginfo.c (choose_hard_reg_mode, record_subregs_of_mode): Likewise.
4184 * regrename.c (build_def_use): Likewise.
4185 * regstat.c (dump_reg_info): Likewise.
4186 * reload.c (complex_word_subreg_p, push_reload, find_dummy_reload)
4187 (find_reloads, find_reloads_subreg_address): Likewise.
4188 * reload1.c (eliminate_regs_1): Likewise.
4189 * rtlanal.c (for_each_inc_dec_find_inc_dec, rtx_cost): Likewise.
4190 * simplify-rtx.c (avoid_constant_pool_reference): Likewise.
4191 (simplify_binary_operation_1, simplify_subreg): Likewise.
4192 * targhooks.c (default_function_arg_padding): Likewise.
4193 (default_hard_regno_nregs, default_class_max_nregs): Likewise.
4194 * tree-cfg.c (verify_gimple_assign_binary): Likewise.
4195 (verify_gimple_assign_ternary): Likewise.
4196 * tree-inline.c (estimate_move_cost): Likewise.
4197 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4198 * tree-ssa-loop-ivopts.c (add_autoinc_candidates): Likewise.
4199 (get_address_cost_ainc): Likewise.
4200 * tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Likewise.
4201 (vect_supportable_dr_alignment): Likewise.
4202 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4203 (vectorizable_reduction): Likewise.
4204 * tree-vect-stmts.c (vectorizable_assignment, vectorizable_shift)
4205 (vectorizable_operation, vectorizable_load): Likewise.
4206 * tree.c (build_same_sized_truth_vector_type): Likewise.
4207 * valtrack.c (cleanup_auto_inc_dec): Likewise.
4208 * var-tracking.c (emit_note_insn_var_location): Likewise.
4209 * config/arc/arc.h (ASM_OUTPUT_CASE_END): Use as_a <scalar_int_mode>.
4210 (ADDR_VEC_ALIGN): Likewise.
4212 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4213 Alan Hayward <alan.hayward@arm.com>
4214 David Sherwood <david.sherwood@arm.com>
4216 * machmode.h (mode_to_bits): Return a poly_uint16 rather than an
4218 (GET_MODE_BITSIZE): Return a constant if ONLY_FIXED_SIZE_MODES,
4219 or if measurement_type is polynomial.
4220 * calls.c (shift_return_value): Treat GET_MODE_BITSIZE as polynomial.
4221 * combine.c (make_extraction): Likewise.
4222 * dse.c (find_shift_sequence): Likewise.
4223 * dwarf2out.c (mem_loc_descriptor): Likewise.
4224 * expmed.c (store_integral_bit_field, extract_bit_field_1): Likewise.
4225 (extract_bit_field, extract_low_bits): Likewise.
4226 * expr.c (convert_move, convert_modes, emit_move_insn_1): Likewise.
4227 (optimize_bitfield_assignment_op, expand_assignment): Likewise.
4228 (store_expr_with_bounds, store_field, expand_expr_real_1): Likewise.
4229 * fold-const.c (optimize_bit_field_compare, merge_ranges): Likewise.
4230 * gimple-fold.c (optimize_atomic_compare_exchange_p): Likewise.
4231 * reload.c (find_reloads): Likewise.
4232 * reload1.c (alter_reg): Likewise.
4233 * stor-layout.c (bitwise_mode_for_mode, compute_record_mode): Likewise.
4234 * targhooks.c (default_secondary_memory_needed_mode): Likewise.
4235 * tree-if-conv.c (predicate_mem_writes): Likewise.
4236 * tree-ssa-strlen.c (handle_builtin_memcmp): Likewise.
4237 * tree-vect-patterns.c (adjust_bool_pattern): Likewise.
4238 * tree-vect-stmts.c (vectorizable_simd_clone_call): Likewise.
4239 * valtrack.c (dead_debug_insert_temp): Likewise.
4240 * varasm.c (mergeable_constant_section): Likewise.
4241 * config/sh/sh.h (LOCAL_ALIGNMENT): Use as_a <fixed_size_mode>.
4243 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4244 Alan Hayward <alan.hayward@arm.com>
4245 David Sherwood <david.sherwood@arm.com>
4247 * expr.c (expand_assignment): Cope with polynomial mode sizes
4248 when assigning to a CONCAT.
4250 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4251 Alan Hayward <alan.hayward@arm.com>
4252 David Sherwood <david.sherwood@arm.com>
4254 * machmode.h (mode_precision): Change from unsigned short to
4256 (mode_to_precision): Return a poly_uint16 rather than an unsigned
4258 (GET_MODE_PRECISION): Return a constant if ONLY_FIXED_SIZE_MODES,
4259 or if measurement_type is not polynomial.
4260 (HWI_COMPUTABLE_MODE_P): Turn into a function. Optimize the case
4261 in which the mode is already known to be a scalar_int_mode.
4262 * genmodes.c (emit_mode_precision): Change the type of mode_precision
4263 from unsigned short to poly_uint16_pod. Use ZERO_COEFFS for the
4265 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4266 for GET_MODE_PRECISION.
4267 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4268 for GET_MODE_PRECISION.
4269 * combine.c (update_rsp_from_reg_equal): Treat GET_MODE_PRECISION
4271 (try_combine, find_split_point, combine_simplify_rtx): Likewise.
4272 (expand_field_assignment, make_extraction): Likewise.
4273 (make_compound_operation_int, record_dead_and_set_regs_1): Likewise.
4274 (get_last_value): Likewise.
4275 * convert.c (convert_to_integer_1): Likewise.
4276 * cse.c (cse_insn): Likewise.
4277 * expr.c (expand_expr_real_1): Likewise.
4278 * lra-constraints.c (simplify_operand_subreg): Likewise.
4279 * optabs-query.c (can_atomic_load_p): Likewise.
4280 * optabs.c (expand_atomic_load): Likewise.
4281 (expand_atomic_store): Likewise.
4282 * ree.c (combine_reaching_defs): Likewise.
4283 * rtl.h (partial_subreg_p, paradoxical_subreg_p): Likewise.
4284 * rtlanal.c (nonzero_bits1, lsb_bitfield_op_p): Likewise.
4285 * tree.h (type_has_mode_precision_p): Likewise.
4286 * ubsan.c (instrument_si_overflow): Likewise.
4288 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4289 Alan Hayward <alan.hayward@arm.com>
4290 David Sherwood <david.sherwood@arm.com>
4292 * tree.h (TYPE_VECTOR_SUBPARTS): Turn into a function and handle
4293 polynomial numbers of units.
4294 (SET_TYPE_VECTOR_SUBPARTS): Likewise.
4295 (valid_vector_subparts_p): New function.
4296 (build_vector_type): Remove temporary shim and take the number
4297 of units as a poly_uint64 rather than an int.
4298 (build_opaque_vector_type): Take the number of units as a
4299 poly_uint64 rather than an int.
4300 * tree.c (build_vector_from_ctor): Handle polynomial
4301 TYPE_VECTOR_SUBPARTS.
4302 (type_hash_canon_hash, type_cache_hasher::equal): Likewise.
4303 (uniform_vector_p, vector_type_mode, build_vector): Likewise.
4304 (build_vector_from_val): If the number of units is variable,
4305 use build_vec_duplicate_cst for constant operands and
4306 VEC_DUPLICATE_EXPR otherwise.
4307 (make_vector_type): Remove temporary is_constant ().
4308 (build_vector_type, build_opaque_vector_type): Take the number of
4309 units as a poly_uint64 rather than an int.
4310 (check_vector_cst): Handle polynomial TYPE_VECTOR_SUBPARTS and
4312 * cfgexpand.c (expand_debug_expr): Likewise.
4313 * expr.c (count_type_elements, categorize_ctor_elements_1): Likewise.
4314 (store_constructor, expand_expr_real_1): Likewise.
4315 (const_scalar_mask_from_tree): Likewise.
4316 * fold-const-call.c (fold_const_reduction): Likewise.
4317 * fold-const.c (const_binop, const_unop, fold_convert_const): Likewise.
4318 (operand_equal_p, fold_vec_perm, fold_ternary_loc): Likewise.
4319 (native_encode_vector, vec_cst_ctor_to_array): Likewise.
4320 (fold_relational_const): Likewise.
4321 (native_interpret_vector): Likewise. Change the size from an
4322 int to an unsigned int.
4323 * gimple-fold.c (gimple_fold_stmt_to_constant_1): Handle polynomial
4324 TYPE_VECTOR_SUBPARTS.
4325 (gimple_fold_indirect_ref, gimple_build_vector): Likewise.
4326 (gimple_build_vector_from_val): Use VEC_DUPLICATE_EXPR when
4327 duplicating a non-constant operand into a variable-length vector.
4328 * hsa-brig.c (hsa_op_immed::emit_to_buffer): Handle polynomial
4329 TYPE_VECTOR_SUBPARTS and VECTOR_CST_NELTS.
4330 * ipa-icf.c (sem_variable::equals): Likewise.
4331 * match.pd: Likewise.
4332 * omp-simd-clone.c (simd_clone_subparts): Likewise.
4333 * print-tree.c (print_node): Likewise.
4334 * stor-layout.c (layout_type): Likewise.
4335 * targhooks.c (default_builtin_vectorization_cost): Likewise.
4336 * tree-cfg.c (verify_gimple_comparison): Likewise.
4337 (verify_gimple_assign_binary): Likewise.
4338 (verify_gimple_assign_ternary): Likewise.
4339 (verify_gimple_assign_single): Likewise.
4340 * tree-pretty-print.c (dump_generic_node): Likewise.
4341 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
4342 (simplify_bitfield_ref, is_combined_permutation_identity): Likewise.
4343 * tree-vect-data-refs.c (vect_permute_store_chain): Likewise.
4344 (vect_grouped_load_supported, vect_permute_load_chain): Likewise.
4345 (vect_shift_permute_load_chain): Likewise.
4346 * tree-vect-generic.c (nunits_for_known_piecewise_op): Likewise.
4347 (expand_vector_condition, optimize_vector_constructor): Likewise.
4348 (lower_vec_perm, get_compute_type): Likewise.
4349 * tree-vect-loop.c (vect_determine_vectorization_factor): Likewise.
4350 (get_initial_defs_for_reduction, vect_transform_loop): Likewise.
4351 * tree-vect-patterns.c (vect_recog_bool_pattern): Likewise.
4352 (vect_recog_mask_conversion_pattern): Likewise.
4353 * tree-vect-slp.c (vect_supported_load_permutation_p): Likewise.
4354 (vect_get_constant_vectors, vect_transform_slp_perm_load): Likewise.
4355 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
4356 (get_group_load_store_type, vectorizable_mask_load_store): Likewise.
4357 (vectorizable_bswap, simd_clone_subparts, vectorizable_assignment)
4358 (vectorizable_shift, vectorizable_operation, vectorizable_store)
4359 (vectorizable_load, vect_is_simple_cond, vectorizable_comparison)
4360 (supportable_widening_operation): Likewise.
4361 (supportable_narrowing_operation): Likewise.
4362 * tree-vector-builder.c (tree_vector_builder::binary_encoded_nelts):
4364 * varasm.c (output_constant): Likewise.
4366 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4367 Alan Hayward <alan.hayward@arm.com>
4368 David Sherwood <david.sherwood@arm.com>
4370 * tree-vect-data-refs.c (vect_permute_store_chain): Reorganize
4371 so that both the length == 3 and length != 3 cases set up their
4372 own permute vectors. Add comments explaining why we know the
4373 number of elements is constant.
4374 (vect_permute_load_chain): Likewise.
4376 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4377 Alan Hayward <alan.hayward@arm.com>
4378 David Sherwood <david.sherwood@arm.com>
4380 * machmode.h (mode_nunits): Change from unsigned char to
4382 (ONLY_FIXED_SIZE_MODES): New macro.
4383 (pod_mode::measurement_type, scalar_int_mode::measurement_type)
4384 (scalar_float_mode::measurement_type, scalar_mode::measurement_type)
4385 (complex_mode::measurement_type, fixed_size_mode::measurement_type):
4387 (mode_to_nunits): Return a poly_uint16 rather than an unsigned short.
4388 (GET_MODE_NUNITS): Return a constant if ONLY_FIXED_SIZE_MODES,
4389 or if measurement_type is not polynomial.
4390 * genmodes.c (ZERO_COEFFS): New macro.
4391 (emit_mode_nunits_inline): Make mode_nunits_inline return a
4393 (emit_mode_nunits): Change the type of mode_nunits to poly_uint16_pod.
4394 Use ZERO_COEFFS when emitting initializers.
4395 * data-streamer.h (bp_pack_poly_value): New function.
4396 (bp_unpack_poly_value): Likewise.
4397 * lto-streamer-in.c (lto_input_mode_table): Use bp_unpack_poly_value
4398 for GET_MODE_NUNITS.
4399 * lto-streamer-out.c (lto_write_mode_table): Use bp_pack_poly_value
4400 for GET_MODE_NUNITS.
4401 * tree.c (make_vector_type): Remove temporary shim and make
4402 the real function take the number of units as a poly_uint64
4404 (build_vector_type_for_mode): Handle polynomial nunits.
4405 * dwarf2out.c (loc_descriptor, add_const_value_attribute): Likewise.
4406 * emit-rtl.c (const_vec_series_p_1): Likewise.
4407 (gen_rtx_CONST_VECTOR): Likewise.
4408 * fold-const.c (test_vec_duplicate_folding): Likewise.
4409 * genrecog.c (validate_pattern): Likewise.
4410 * optabs-query.c (can_vec_perm_var_p, can_mult_highpart_p): Likewise.
4411 * optabs-tree.c (expand_vec_cond_expr_p): Likewise.
4412 * optabs.c (expand_vector_broadcast, expand_binop_directly): Likewise.
4413 (shift_amt_for_vec_perm_mask, expand_vec_perm_var): Likewise.
4414 (expand_vec_cond_expr, expand_mult_highpart): Likewise.
4415 * rtlanal.c (subreg_get_info): Likewise.
4416 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
4417 (vect_grouped_load_supported): Likewise.
4418 * tree-vect-generic.c (type_for_widest_vector_mode): Likewise.
4419 * tree-vect-loop.c (have_whole_vector_shift): Likewise.
4420 * simplify-rtx.c (simplify_unary_operation_1): Likewise.
4421 (simplify_const_unary_operation, simplify_binary_operation_1)
4422 (simplify_const_binary_operation, simplify_ternary_operation)
4423 (test_vector_ops_duplicate, test_vector_ops): Likewise.
4424 (simplify_immed_subreg): Use GET_MODE_NUNITS on a fixed_size_mode
4425 instead of CONST_VECTOR_NUNITS.
4426 * varasm.c (output_constant_pool_2): Likewise.
4427 * rtx-vector-builder.c (rtx_vector_builder::build): Only include the
4428 explicit-encoded elements in the XVEC for variable-length vectors.
4430 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4432 * lra-constraints.c (curr_insn_transform): Use partial_subreg_p.
4434 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4435 Alan Hayward <alan.hayward@arm.com>
4436 David Sherwood <david.sherwood@arm.com>
4438 * coretypes.h (fixed_size_mode): Declare.
4439 (fixed_size_mode_pod): New typedef.
4440 * builtins.h (target_builtins::x_apply_args_mode)
4441 (target_builtins::x_apply_result_mode): Change type to
4442 fixed_size_mode_pod.
4443 * builtins.c (apply_args_size, apply_result_size, result_vector)
4444 (expand_builtin_apply_args_1, expand_builtin_apply)
4445 (expand_builtin_return): Update accordingly.
4447 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4449 * cse.c (hash_rtx_cb): Hash only the encoded elements.
4450 * cselib.c (cselib_hash_rtx): Likewise.
4451 * expmed.c (make_tree): Build VECTOR_CSTs directly from the
4452 CONST_VECTOR encoding.
4454 2017-01-03 Jakub Jelinek <jakub@redhat.com>
4455 Jeff Law <law@redhat.com>
4458 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): For
4459 noreturn probe, use gen_pop instead of ix86_emit_restore_reg_using_pop,
4460 only set RTX_FRAME_RELATED_P on both the push and pop if cfa_reg is sp
4461 and add REG_CFA_ADJUST_CFA notes in that case to both insns.
4464 * config/i386/i386.c (ix86_adjust_stack_and_probe_stack_clash): Do not
4465 explicitly probe *sp in a noreturn function if there were any callee
4466 register saves or frame pointer is needed.
4468 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4471 * cfgexpand.c (expand_debug_expr): Return NULL if mode is
4472 BLKmode for ternary, binary or unary expressions.
4475 * var-tracking.c (delete_vta_debug_insn): New inline function.
4476 (delete_vta_debug_insns): Add USE_CFG argument, if true, walk just
4477 insns from get_insns () to NULL instead of each bb separately.
4478 Use delete_vta_debug_insn. No longer static.
4479 (vt_debug_insns_local, variable_tracking_main_1): Adjust
4480 delete_vta_debug_insns callers.
4481 * rtl.h (delete_vta_debug_insns): Declare.
4482 * final.c (rest_of_handle_final): Call delete_vta_debug_insns
4483 instead of variable_tracking_main.
4485 2018-01-03 Martin Sebor <msebor@redhat.com>
4487 PR tree-optimization/83603
4488 * calls.c (maybe_warn_nonstring_arg): Avoid accessing function
4489 arguments past the endof the argument list in functions declared
4490 without a prototype.
4491 * gimple-ssa-warn-restrict.c (wrestrict_dom_walker::check_call):
4492 Avoid checking when arguments are null.
4494 2018-01-03 Martin Sebor <msebor@redhat.com>
4497 * doc/extend.texi (attribute const): Fix a typo.
4498 * ipa-pure-const.c ((warn_function_const, warn_function_pure): Avoid
4499 issuing -Wsuggest-attribute for void functions.
4501 2018-01-03 Martin Sebor <msebor@redhat.com>
4503 * gimple-ssa-warn-restrict.c (builtin_memref::builtin_memref): Use
4504 offset_int::from instead of wide_int::to_shwi.
4505 (maybe_diag_overlap): Remove assertion.
4506 Use HOST_WIDE_INT_PRINT_DEC instead of %lli.
4507 * gimple-ssa-sprintf.c (format_directive): Same.
4508 (parse_directive): Same.
4509 (sprintf_dom_walker::compute_format_length): Same.
4510 (try_substitute_return_value): Same.
4512 2017-01-03 Jeff Law <law@redhat.com>
4515 * explow.c (anti_adjust_stack_and_probe_stack_clash): Test a
4516 non-constant residual for zero at runtime and avoid probing in
4517 that case. Reorganize code for trailing problem to mirror handling
4520 2018-01-03 Prathamesh Kulkarni <prathamesh.kulkarni@linaro.org>
4522 PR tree-optimization/83501
4523 * tree-ssa-strlen.c (get_string_cst): New.
4524 (handle_char_store): Call get_string_cst.
4526 2018-01-03 Martin Liska <mliska@suse.cz>
4528 PR tree-optimization/83593
4529 * tree-ssa-strlen.c: Include tree-cfg.h.
4530 (strlen_check_and_optimize_stmt): Add new argument cleanup_eh.
4531 (strlen_dom_walker): Add new member variable m_cleanup_cfg.
4532 (strlen_dom_walker::strlen_dom_walker): Initialize m_cleanup_cfg
4534 (strlen_dom_walker::before_dom_children): Call
4535 gimple_purge_dead_eh_edges. Dump tranformation with details
4537 (strlen_dom_walker::before_dom_children): Update call by adding
4538 new argument cleanup_eh.
4539 (pass_strlen::execute): Return TODO_cleanup_cfg if needed.
4541 2018-01-03 Martin Liska <mliska@suse.cz>
4544 * cif-code.def (VARIADIC_THUNK): New enum value.
4545 * ipa-fnsummary.c (compute_fn_summary): Do not inline variadic
4548 2018-01-03 Jan Beulich <jbeulich@suse.com>
4550 * sse.md (mov<mode>_internal): Tighten condition for when to use
4551 vmovdqu<ssescalarsize> for TI and OI modes.
4553 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4555 Update copyright years.
4557 2018-01-03 Martin Liska <mliska@suse.cz>
4560 * ipa-visibility.c (function_and_variable_visibility): Skip
4561 functions with noipa attribure.
4563 2018-01-03 Jakub Jelinek <jakub@redhat.com>
4565 * gcc.c (process_command): Update copyright notice dates.
4566 * gcov-dump.c (print_version): Ditto.
4567 * gcov.c (print_version): Ditto.
4568 * gcov-tool.c (print_version): Ditto.
4569 * gengtype.c (create_file): Ditto.
4570 * doc/cpp.texi: Bump @copying's copyright year.
4571 * doc/cppinternals.texi: Ditto.
4572 * doc/gcc.texi: Ditto.
4573 * doc/gccint.texi: Ditto.
4574 * doc/gcov.texi: Ditto.
4575 * doc/install.texi: Ditto.
4576 * doc/invoke.texi: Ditto.
4578 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4580 * vector-builder.h (vector_builder::m_full_nelts): Change from
4581 unsigned int to poly_uint64.
4582 (vector_builder::full_nelts): Update prototype accordingly.
4583 (vector_builder::new_vector): Likewise.
4584 (vector_builder::encoded_full_vector_p): Handle polynomial full_nelts.
4585 (vector_builder::operator ==): Likewise.
4586 (vector_builder::finalize): Likewise.
4587 * int-vector-builder.h (int_vector_builder::int_vector_builder):
4588 Take the number of elements as a poly_uint64 rather than an
4590 * vec-perm-indices.h (vec_perm_indices::m_nelts_per_input): Change
4591 from unsigned int to poly_uint64.
4592 (vec_perm_indices::vec_perm_indices): Update prototype accordingly.
4593 (vec_perm_indices::new_vector): Likewise.
4594 (vec_perm_indices::length): Likewise.
4595 (vec_perm_indices::nelts_per_input): Likewise.
4596 (vec_perm_indices::input_nelts): Likewise.
4597 * vec-perm-indices.c (vec_perm_indices::new_vector): Take the
4598 number of elements per input as a poly_uint64 rather than an
4599 unsigned int. Use the original encoding for variable-length
4600 vectors, rather than clamping each individual element.
4601 For the second and subsequent elements in each pattern,
4602 clamp the step and base before clamping their sum.
4603 (vec_perm_indices::series_p): Handle polynomial element counts.
4604 (vec_perm_indices::all_in_range_p): Likewise.
4605 (vec_perm_indices_to_tree): Likewise.
4606 (vec_perm_indices_to_rtx): Likewise.
4607 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise.
4608 * tree-vector-builder.c (tree_vector_builder::new_unary_operation)
4609 (tree_vector_builder::new_binary_operation): Handle polynomial
4610 element counts. Return false if we need to know the number
4611 of elements at compile time.
4612 * fold-const.c (fold_vec_perm): Punt if the number of elements
4613 isn't known at compile time.
4615 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4617 * vec-perm-indices.h (vec_perm_builder): Change element type
4618 from HOST_WIDE_INT to poly_int64.
4619 (vec_perm_indices::element_type): Update accordingly.
4620 (vec_perm_indices::clamp): Handle polynomial element_types.
4621 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
4622 (vec_perm_indices::all_in_range_p): Likewise.
4623 (tree_to_vec_perm_builder): Check for poly_int64 trees rather
4625 * vector-builder.h (vector_builder::stepped_sequence_p): Handle
4626 polynomial vec_perm_indices element types.
4627 * int-vector-builder.h (int_vector_builder::equal_p): Likewise.
4628 * fold-const.c (fold_vec_perm): Likewise.
4629 * optabs.c (shift_amt_for_vec_perm_mask): Likewise.
4630 * tree-vect-generic.c (lower_vec_perm): Likewise.
4631 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
4632 * config/aarch64/aarch64.c (aarch64_evpc_tbl): Cast d->perm
4633 element type to HOST_WIDE_INT.
4635 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4636 Alan Hayward <alan.hayward@arm.com>
4637 David Sherwood <david.sherwood@arm.com>
4639 * alias.c (addr_side_effect_eval): Take the size as a poly_int64
4640 rather than an int. Use plus_constant.
4641 (memrefs_conflict_p): Take the sizes as poly_int64s rather than ints.
4642 Take the offset "c" as a poly_int64 rather than a HOST_WIDE_INT.
4644 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4645 Alan Hayward <alan.hayward@arm.com>
4646 David Sherwood <david.sherwood@arm.com>
4648 * calls.c (emit_call_1, expand_call): Change struct_value_size from
4649 a HOST_WIDE_INT to a poly_int64.
4651 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4652 Alan Hayward <alan.hayward@arm.com>
4653 David Sherwood <david.sherwood@arm.com>
4655 * calls.c (load_register_parameters): Cope with polynomial
4656 mode sizes. Require a constant size for BLKmode parameters
4657 that aren't described by a PARALLEL. If BLOCK_REG_PADDING
4658 forces a parameter to be padded at the lsb end in order to
4659 fill a complete number of words, require the parameter size
4660 to be ordered wrt UNITS_PER_WORD.
4662 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4663 Alan Hayward <alan.hayward@arm.com>
4664 David Sherwood <david.sherwood@arm.com>
4666 * reload1.c (spill_stack_slot_width): Change element type
4667 from unsigned int to poly_uint64_pod.
4668 (alter_reg): Treat mode sizes as polynomial.
4670 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4671 Alan Hayward <alan.hayward@arm.com>
4672 David Sherwood <david.sherwood@arm.com>
4674 * reload.c (complex_word_subreg_p): New function.
4675 (reload_inner_reg_of_subreg, push_reload): Use it.
4677 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4678 Alan Hayward <alan.hayward@arm.com>
4679 David Sherwood <david.sherwood@arm.com>
4681 * lra-constraints.c (process_alt_operands): Reject matched
4682 operands whose sizes aren't ordered.
4683 (match_reload): Refer to this check here.
4685 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4686 Alan Hayward <alan.hayward@arm.com>
4687 David Sherwood <david.sherwood@arm.com>
4689 * builtins.c (expand_ifn_atomic_compare_exchange_into_call): Assert
4690 that the mode size is in the set {1, 2, 4, 8, 16}.
4692 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4693 Alan Hayward <alan.hayward@arm.com>
4694 David Sherwood <david.sherwood@arm.com>
4696 * var-tracking.c (adjust_mems): Treat mode sizes as polynomial.
4697 Use plus_constant instead of gen_rtx_PLUS.
4699 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4700 Alan Hayward <alan.hayward@arm.com>
4701 David Sherwood <david.sherwood@arm.com>
4703 * config/cr16/cr16-protos.h (cr16_push_rounding): Declare.
4704 * config/cr16/cr16.h (PUSH_ROUNDING): Move implementation to...
4705 * config/cr16/cr16.c (cr16_push_rounding): ...this new function.
4706 * config/h8300/h8300-protos.h (h8300_push_rounding): Declare.
4707 * config/h8300/h8300.h (PUSH_ROUNDING): Move implementation to...
4708 * config/h8300/h8300.c (h8300_push_rounding): ...this new function.
4709 * config/i386/i386-protos.h (ix86_push_rounding): Declare.
4710 * config/i386/i386.h (PUSH_ROUNDING): Move implementation to...
4711 * config/i386/i386.c (ix86_push_rounding): ...this new function.
4712 * config/m32c/m32c-protos.h (m32c_push_rounding): Take and return
4714 * config/m32c/m32c.c (m32c_push_rounding): Likewise.
4715 * config/m68k/m68k-protos.h (m68k_push_rounding): Declare.
4716 * config/m68k/m68k.h (PUSH_ROUNDING): Move implementation to...
4717 * config/m68k/m68k.c (m68k_push_rounding): ...this new function.
4718 * config/pdp11/pdp11-protos.h (pdp11_push_rounding): Declare.
4719 * config/pdp11/pdp11.h (PUSH_ROUNDING): Move implementation to...
4720 * config/pdp11/pdp11.c (pdp11_push_rounding): ...this new function.
4721 * config/stormy16/stormy16-protos.h (xstormy16_push_rounding): Declare.
4722 * config/stormy16/stormy16.h (PUSH_ROUNDING): Move implementation to...
4723 * config/stormy16/stormy16.c (xstormy16_push_rounding): ...this new
4725 * expr.c (emit_move_resolve_push): Treat the input and result
4726 of PUSH_ROUNDING as a poly_int64.
4727 (emit_move_complex_push, emit_single_push_insn_1): Likewise.
4728 (emit_push_insn): Likewise.
4729 * lra-eliminations.c (mark_not_eliminable): Likewise.
4730 * recog.c (push_operand): Likewise.
4731 * reload1.c (elimination_effects): Likewise.
4732 * rtlanal.c (nonzero_bits1): Likewise.
4733 * calls.c (store_one_arg): Likewise. Require the padding to be
4734 known at compile time.
4736 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4737 Alan Hayward <alan.hayward@arm.com>
4738 David Sherwood <david.sherwood@arm.com>
4740 * expr.c (emit_single_push_insn_1): Treat mode sizes as polynomial.
4741 Use plus_constant instead of gen_rtx_PLUS.
4743 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4744 Alan Hayward <alan.hayward@arm.com>
4745 David Sherwood <david.sherwood@arm.com>
4747 * auto-inc-dec.c (set_inc_state): Take the mode size as a poly_int64
4750 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4751 Alan Hayward <alan.hayward@arm.com>
4752 David Sherwood <david.sherwood@arm.com>
4754 * expr.c (expand_expr_real_1): Use tree_to_poly_uint64
4755 instead of int_size_in_bytes when handling VIEW_CONVERT_EXPRs
4756 via stack temporaries. Treat the mode size as polynomial too.
4758 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4759 Alan Hayward <alan.hayward@arm.com>
4760 David Sherwood <david.sherwood@arm.com>
4762 * expr.c (expand_expr_real_2): When handling conversions involving
4763 unions, apply tree_to_poly_uint64 to the TYPE_SIZE rather than
4764 multiplying int_size_in_bytes by BITS_PER_UNIT. Treat GET_MODE_BISIZE
4765 as a poly_uint64 too.
4767 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4768 Alan Hayward <alan.hayward@arm.com>
4769 David Sherwood <david.sherwood@arm.com>
4771 * rtlanal.c (subreg_get_info): Handle polynomial mode sizes.
4773 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4774 Alan Hayward <alan.hayward@arm.com>
4775 David Sherwood <david.sherwood@arm.com>
4777 * combine.c (can_change_dest_mode): Handle polynomial
4778 REGMODE_NATURAL_SIZE.
4779 * expmed.c (store_bit_field_1): Likewise.
4780 * expr.c (store_constructor): Likewise.
4781 * emit-rtl.c (validate_subreg): Operate on polynomial mode sizes
4782 and polynomial REGMODE_NATURAL_SIZE.
4783 (gen_lowpart_common): Likewise.
4784 * reginfo.c (record_subregs_of_mode): Likewise.
4785 * rtlanal.c (read_modify_subreg_p): Likewise.
4787 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4788 Alan Hayward <alan.hayward@arm.com>
4789 David Sherwood <david.sherwood@arm.com>
4791 * internal-fn.c (expand_vector_ubsan_overflow): Handle polynomial
4792 numbers of elements.
4794 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4795 Alan Hayward <alan.hayward@arm.com>
4796 David Sherwood <david.sherwood@arm.com>
4798 * match.pd: Cope with polynomial numbers of vector elements.
4800 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4801 Alan Hayward <alan.hayward@arm.com>
4802 David Sherwood <david.sherwood@arm.com>
4804 * fold-const.c (fold_indirect_ref_1): Handle polynomial offsets
4805 in a POINTER_PLUS_EXPR.
4807 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4808 Alan Hayward <alan.hayward@arm.com>
4809 David Sherwood <david.sherwood@arm.com>
4811 * omp-simd-clone.c (simd_clone_subparts): New function.
4812 (simd_clone_init_simd_arrays): Use it instead of TYPE_VECTOR_SUBPARTS.
4813 (ipa_simd_modify_function_body): Likewise.
4815 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4816 Alan Hayward <alan.hayward@arm.com>
4817 David Sherwood <david.sherwood@arm.com>
4819 * tree-vect-generic.c (nunits_for_known_piecewise_op): New function.
4820 (expand_vector_piecewise): Use it instead of TYPE_VECTOR_SUBPARTS.
4821 (expand_vector_addition, add_rshift, expand_vector_divmod): Likewise.
4822 (expand_vector_condition, vector_element): Likewise.
4823 (subparts_gt): New function.
4824 (get_compute_type): Use subparts_gt.
4825 (count_type_subparts): Delete.
4826 (expand_vector_operations_1): Use subparts_gt instead of
4827 count_type_subparts.
4829 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4830 Alan Hayward <alan.hayward@arm.com>
4831 David Sherwood <david.sherwood@arm.com>
4833 * tree-vect-data-refs.c (vect_no_alias_p): Replace with...
4834 (vect_compile_time_alias): ...this new function. Do the calculation
4835 on poly_ints rather than trees.
4836 (vect_prune_runtime_alias_test_list): Update call accordingly.
4838 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4839 Alan Hayward <alan.hayward@arm.com>
4840 David Sherwood <david.sherwood@arm.com>
4842 * tree-vect-slp.c (vect_build_slp_tree_1): Handle polynomial
4844 (vect_schedule_slp_instance): Likewise.
4846 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4847 Alan Hayward <alan.hayward@arm.com>
4848 David Sherwood <david.sherwood@arm.com>
4850 * tree-vect-slp.c (vect_get_and_check_slp_defs): Reject
4851 constant and extern definitions for variable-length vectors.
4852 (vect_get_constant_vectors): Note that the number of units
4853 is known to be constant.
4855 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4856 Alan Hayward <alan.hayward@arm.com>
4857 David Sherwood <david.sherwood@arm.com>
4859 * tree-vect-stmts.c (vectorizable_conversion): Treat the number
4860 of units as polynomial. Choose between WIDE and NARROW based
4863 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4864 Alan Hayward <alan.hayward@arm.com>
4865 David Sherwood <david.sherwood@arm.com>
4867 * tree-vect-stmts.c (simd_clone_subparts): New function.
4868 (vectorizable_simd_clone_call): Use it instead of TYPE_VECTOR_SUBPARTS.
4870 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4871 Alan Hayward <alan.hayward@arm.com>
4872 David Sherwood <david.sherwood@arm.com>
4874 * tree-vect-stmts.c (vectorizable_call): Treat the number of
4875 vectors as polynomial. Use build_index_vector for
4878 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4879 Alan Hayward <alan.hayward@arm.com>
4880 David Sherwood <david.sherwood@arm.com>
4882 * tree-vect-stmts.c (get_load_store_type): Treat the number of
4883 units as polynomial. Reject VMAT_ELEMENTWISE and VMAT_STRIDED_SLP
4884 for variable-length vectors.
4885 (vectorizable_mask_load_store): Treat the number of units as
4886 polynomial, asserting that it is constant if the condition has
4887 already been enforced.
4888 (vectorizable_store, vectorizable_load): Likewise.
4890 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4891 Alan Hayward <alan.hayward@arm.com>
4892 David Sherwood <david.sherwood@arm.com>
4894 * tree-vect-loop.c (vectorizable_live_operation): Treat the number
4895 of units as polynomial. Punt if we can't tell at compile time
4896 which vector contains the final result.
4898 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4899 Alan Hayward <alan.hayward@arm.com>
4900 David Sherwood <david.sherwood@arm.com>
4902 * tree-vect-loop.c (vectorizable_induction): Treat the number
4903 of units as polynomial. Punt on SLP inductions. Use an integer
4904 VEC_SERIES_EXPR for variable-length integer reductions. Use a
4905 cast of such a series for variable-length floating-point
4908 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4909 Alan Hayward <alan.hayward@arm.com>
4910 David Sherwood <david.sherwood@arm.com>
4912 * tree.h (build_index_vector): Declare.
4913 * tree.c (build_index_vector): New function.
4914 * tree-vect-loop.c (get_initial_defs_for_reduction): Treat the number
4915 of units as polynomial, forcibly converting it to a constant if
4916 vectorizable_reduction has already enforced the condition.
4917 (vect_create_epilog_for_reduction): Likewise. Use build_index_vector
4918 to create a {1,2,3,...} vector.
4919 (vectorizable_reduction): Treat the number of units as polynomial.
4920 Choose vectype_in based on the largest scalar element size rather
4921 than the smallest number of units. Enforce the restrictions
4924 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4925 Alan Hayward <alan.hayward@arm.com>
4926 David Sherwood <david.sherwood@arm.com>
4928 * tree-vect-data-refs.c (vector_alignment_reachable_p): Treat the
4929 number of units as polynomial.
4931 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4932 Alan Hayward <alan.hayward@arm.com>
4933 David Sherwood <david.sherwood@arm.com>
4935 * target.h (vector_sizes, auto_vector_sizes): New typedefs.
4936 * target.def (autovectorize_vector_sizes): Return the vector sizes
4937 by pointer, using vector_sizes rather than a bitmask.
4938 * targhooks.h (default_autovectorize_vector_sizes): Update accordingly.
4939 * targhooks.c (default_autovectorize_vector_sizes): Likewise.
4940 * config/aarch64/aarch64.c (aarch64_autovectorize_vector_sizes):
4942 * config/arc/arc.c (arc_autovectorize_vector_sizes): Likewise.
4943 * config/arm/arm.c (arm_autovectorize_vector_sizes): Likewise.
4944 * config/i386/i386.c (ix86_autovectorize_vector_sizes): Likewise.
4945 * config/mips/mips.c (mips_autovectorize_vector_sizes): Likewise.
4946 * omp-general.c (omp_max_vf): Likewise.
4947 * omp-low.c (omp_clause_aligned_alignment): Likewise.
4948 * optabs-query.c (can_vec_mask_load_store_p): Likewise.
4949 * tree-vect-loop.c (vect_analyze_loop): Likewise.
4950 * tree-vect-slp.c (vect_slp_bb): Likewise.
4951 * doc/tm.texi: Regenerate.
4952 * tree-vectorizer.h (current_vector_size): Change from an unsigned int
4954 * tree-vect-stmts.c (get_vectype_for_scalar_type_and_size): Take
4955 the vector size as a poly_uint64 rather than an unsigned int.
4956 (current_vector_size): Change from an unsigned int to a poly_uint64.
4957 (get_vectype_for_scalar_type): Update accordingly.
4958 * tree.h (build_truth_vector_type): Take the size and number of
4959 units as a poly_uint64 rather than an unsigned int.
4960 (build_vector_type): Add a temporary overload that takes
4961 the number of units as a poly_uint64 rather than an unsigned int.
4962 * tree.c (make_vector_type): Likewise.
4963 (build_truth_vector_type): Take the number of units as a poly_uint64
4964 rather than an unsigned int.
4966 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4967 Alan Hayward <alan.hayward@arm.com>
4968 David Sherwood <david.sherwood@arm.com>
4970 * target.def (get_mask_mode): Take the number of units and length
4971 as poly_uint64s rather than unsigned ints.
4972 * targhooks.h (default_get_mask_mode): Update accordingly.
4973 * targhooks.c (default_get_mask_mode): Likewise.
4974 * config/i386/i386.c (ix86_get_mask_mode): Likewise.
4975 * doc/tm.texi: Regenerate.
4977 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4978 Alan Hayward <alan.hayward@arm.com>
4979 David Sherwood <david.sherwood@arm.com>
4981 * omp-general.h (omp_max_vf): Return a poly_uint64 instead of an int.
4982 * omp-general.c (omp_max_vf): Likewise.
4983 * omp-expand.c (omp_adjust_chunk_size): Update call to omp_max_vf.
4984 (expand_omp_simd): Handle polynomial safelen.
4985 * omp-low.c (omplow_simd_context): Add a default constructor.
4986 (omplow_simd_context::max_vf): Change from int to poly_uint64.
4987 (lower_rec_simd_input_clauses): Update accordingly.
4988 (lower_rec_input_clauses): Likewise.
4990 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
4991 Alan Hayward <alan.hayward@arm.com>
4992 David Sherwood <david.sherwood@arm.com>
4994 * tree-vectorizer.h (vect_nunits_for_cost): New function.
4995 * tree-vect-loop.c (vect_model_reduction_cost): Use it.
4996 * tree-vect-slp.c (vect_analyze_slp_cost_1): Likewise.
4997 (vect_analyze_slp_cost): Likewise.
4998 * tree-vect-stmts.c (vect_model_store_cost): Likewise.
4999 (vect_model_load_cost): Likewise.
5001 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5002 Alan Hayward <alan.hayward@arm.com>
5003 David Sherwood <david.sherwood@arm.com>
5005 * tree-vect-slp.c (vect_record_max_nunits, vect_build_slp_tree_1)
5006 (vect_build_slp_tree_2, vect_build_slp_tree): Change max_nunits
5007 from an unsigned int * to a poly_uint64_pod *.
5008 (calculate_unrolling_factor): New function.
5009 (vect_analyze_slp_instance): Use it. Track polynomial max_nunits.
5011 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5012 Alan Hayward <alan.hayward@arm.com>
5013 David Sherwood <david.sherwood@arm.com>
5015 * tree-vectorizer.h (_slp_instance::unrolling_factor): Change
5016 from an unsigned int to a poly_uint64.
5017 (_loop_vec_info::slp_unrolling_factor): Likewise.
5018 (_loop_vec_info::vectorization_factor): Change from an int
5020 (MAX_VECTORIZATION_FACTOR): Bump from 64 to INT_MAX.
5021 (vect_get_num_vectors): New function.
5022 (vect_update_max_nunits, vect_vf_for_cost): Likewise.
5023 (vect_get_num_copies): Use vect_get_num_vectors.
5024 (vect_analyze_data_ref_dependences): Change max_vf from an int *
5025 to an unsigned int *.
5026 (vect_analyze_data_refs): Change min_vf from an int * to a
5028 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5029 than an unsigned HOST_WIDE_INT.
5030 * tree-vect-data-refs.c (vect_analyze_possibly_independent_ddr)
5031 (vect_analyze_data_ref_dependence): Change max_vf from an int *
5032 to an unsigned int *.
5033 (vect_analyze_data_ref_dependences): Likewise.
5034 (vect_compute_data_ref_alignment): Handle polynomial vf.
5035 (vect_enhance_data_refs_alignment): Likewise.
5036 (vect_prune_runtime_alias_test_list): Likewise.
5037 (vect_shift_permute_load_chain): Likewise.
5038 (vect_supportable_dr_alignment): Likewise.
5039 (dependence_distance_ge_vf): Take the vectorization factor as a
5040 poly_uint64 rather than an unsigned HOST_WIDE_INT.
5041 (vect_analyze_data_refs): Change min_vf from an int * to a
5043 * tree-vect-loop-manip.c (vect_gen_scalar_loop_niters): Take
5044 vfm1 as a poly_uint64 rather than an int. Make the same change
5045 for the returned bound_scalar.
5046 (vect_gen_vector_loop_niters): Handle polynomial vf.
5047 (vect_do_peeling): Likewise. Update call to
5048 vect_gen_scalar_loop_niters and handle polynomial bound_scalars.
5049 (vect_gen_vector_loop_niters_mult_vf): Assert that the vf must
5051 * tree-vect-loop.c (vect_determine_vectorization_factor)
5052 (vect_update_vf_for_slp, vect_analyze_loop_2): Handle polynomial vf.
5053 (vect_get_known_peeling_cost): Likewise.
5054 (vect_estimate_min_profitable_iters, vectorizable_reduction): Likewise.
5055 (vect_worthwhile_without_simd_p, vectorizable_induction): Likewise.
5056 (vect_transform_loop): Likewise. Use the lowest possible VF when
5057 updating the upper bounds of the loop.
5058 (vect_min_worthwhile_factor): Make static. Return an unsigned int
5060 * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts): Cope with
5061 polynomial unroll factors.
5062 (vect_analyze_slp_cost_1, vect_analyze_slp_instance): Likewise.
5063 (vect_make_slp_decision): Likewise.
5064 (vect_supported_load_permutation_p): Likewise, and polynomial
5066 (vect_analyze_slp_cost): Handle polynomial vf.
5067 (vect_slp_analyze_node_operations): Likewise.
5068 (vect_slp_analyze_bb_1): Likewise.
5069 (vect_transform_slp_perm_load): Take the vf as a poly_uint64 rather
5070 than an unsigned HOST_WIDE_INT.
5071 * tree-vect-stmts.c (vectorizable_simd_clone_call, vectorizable_store)
5072 (vectorizable_load): Handle polynomial vf.
5073 * tree-vectorizer.c (simduid_to_vf::vf): Change from an int to
5075 (adjust_simduid_builtins, shrink_simd_arrays): Update accordingly.
5077 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5078 Alan Hayward <alan.hayward@arm.com>
5079 David Sherwood <david.sherwood@arm.com>
5081 * match.pd: Handle bit operations involving three constants
5082 and try to fold one pair.
5084 2018-01-03 Richard Sandiford <richard.sandiford@linaro.org>
5086 * tree-vect-loop-manip.c: Include gimple-fold.h.
5087 (slpeel_make_loop_iterate_ntimes): Add step, final_iv and
5088 niters_maybe_zero parameters. Handle other cases besides a step of 1.
5089 (vect_gen_vector_loop_niters): Add a step_vector_ptr parameter.
5090 Add a path that uses a step of VF instead of 1, but disable it
5092 (vect_do_peeling): Add step_vector, niters_vector_mult_vf_var
5093 and niters_no_overflow parameters. Update calls to
5094 slpeel_make_loop_iterate_ntimes and vect_gen_vector_loop_niters.
5095 Create a new SSA name if the latter choses to use a ste other
5096 than zero, and return it via niters_vector_mult_vf_var.
5097 * tree-vect-loop.c (vect_transform_loop): Update calls to
5098 vect_do_peeling, vect_gen_vector_loop_niters and
5099 slpeel_make_loop_iterate_ntimes.
5100 * tree-vectorizer.h (slpeel_make_loop_iterate_ntimes, vect_do_peeling)
5101 (vect_gen_vector_loop_niters): Update declarations after above changes.
5103 2018-01-02 Michael Meissner <meissner@linux.vnet.ibm.com>
5105 * config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
5106 128-bit round to integer instructions.
5107 (ceil<mode>2): Likewise.
5108 (btrunc<mode>2): Likewise.
5109 (round<mode>2): Likewise.
5111 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5113 * config/rs6000/rs6000-string.c (expand_block_move): Allow the use of
5114 unaligned VSX load/store on P8/P9.
5115 (expand_block_clear): Allow the use of unaligned VSX
5116 load/store on P8/P9.
5118 2018-01-02 Bill Schmidt <wschmidt@linux.vnet.ibm.com>
5120 * config/rs6000/rs6000-p8swap.c (swap_feeds_both_load_and_store):
5122 (rs6000_analyze_swaps): Mark a web unoptimizable if it contains a
5123 swap associated with both a load and a store.
5125 2018-01-02 Andrew Waterman <andrew@sifive.com>
5127 * config/riscv/linux.h (ICACHE_FLUSH_FUNC): New.
5128 * config/riscv/riscv.md (clear_cache): Use it.
5130 2018-01-02 Artyom Skrobov <tyomitch@gmail.com>
5132 * web.c: Remove out-of-date comment.
5134 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5136 * expr.c (fixup_args_size_notes): Check that any existing
5137 REG_ARGS_SIZE notes are correct, and don't try to re-add them.
5138 (emit_single_push_insn_1): Move stack_pointer_delta adjustment to...
5139 (emit_single_push_insn): ...here.
5141 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5143 * rtl.h (CONST_VECTOR_ELT): Redefine to const_vector_elt.
5144 (const_vector_encoded_nelts): New function.
5145 (CONST_VECTOR_NUNITS): Redefine to use GET_MODE_NUNITS.
5146 (const_vector_int_elt, const_vector_elt): Declare.
5147 * emit-rtl.c (const_vector_int_elt_1): New function.
5148 (const_vector_elt): Likewise.
5149 * simplify-rtx.c (simplify_immed_subreg): Avoid taking the address
5150 of CONST_VECTOR_ELT.
5152 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5154 * expr.c: Include rtx-vector-builder.h.
5155 (const_vector_mask_from_tree): Use rtx_vector_builder and operate
5156 directly on the tree encoding.
5157 (const_vector_from_tree): Likewise.
5158 * optabs.c: Include rtx-vector-builder.h.
5159 (expand_vec_perm_var): Use rtx_vector_builder and create a repeating
5160 sequence of "u" values.
5161 * vec-perm-indices.c: Include rtx-vector-builder.h.
5162 (vec_perm_indices_to_rtx): Use rtx_vector_builder and operate
5163 directly on the vec_perm_indices encoding.
5165 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5167 * doc/rtl.texi (const_vector): Describe new encoding scheme.
5168 * Makefile.in (OBJS): Add rtx-vector-builder.o.
5169 * rtx-vector-builder.h: New file.
5170 * rtx-vector-builder.c: Likewise.
5171 * rtl.h (rtx_def::u2): Add a const_vector field.
5172 (CONST_VECTOR_NPATTERNS): New macro.
5173 (CONST_VECTOR_NELTS_PER_PATTERN): Likewise.
5174 (CONST_VECTOR_DUPLICATE_P): Likewise.
5175 (CONST_VECTOR_STEPPED_P): Likewise.
5176 (CONST_VECTOR_ENCODED_ELT): Likewise.
5177 (const_vec_duplicate_p): Check for a duplicated vector encoding.
5178 (unwrap_const_vec_duplicate): Likewise.
5179 (const_vec_series_p): Check for a non-duplicated vector encoding.
5180 Say that the function only returns true for integer vectors.
5181 * emit-rtl.c: Include rtx-vector-builder.h.
5182 (gen_const_vec_duplicate_1): Delete.
5183 (gen_const_vector): Call gen_const_vec_duplicate instead of
5184 gen_const_vec_duplicate_1.
5185 (const_vec_series_p_1): Operate directly on the CONST_VECTOR encoding.
5186 (gen_const_vec_duplicate): Use rtx_vector_builder.
5187 (gen_const_vec_series): Likewise.
5188 (gen_rtx_CONST_VECTOR): Likewise.
5189 * config/powerpcspe/powerpcspe.c: Include rtx-vector-builder.h.
5190 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5191 Build a new vector rather than modifying a CONST_VECTOR in-place.
5192 (handle_special_swappables): Update call accordingly.
5193 * config/rs6000/rs6000-p8swap.c: Include rtx-vector-builder.h.
5194 (swap_const_vector_halves): Take an rtx pointer rather than rtx.
5195 Build a new vector rather than modifying a CONST_VECTOR in-place.
5196 (handle_special_swappables): Update call accordingly.
5198 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5200 * simplify-rtx.c (simplify_const_binary_operation): Use
5201 CONST_VECTOR_ELT instead of XVECEXP.
5203 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5205 * tree-cfg.c (verify_gimple_assign_ternary): Allow the size of
5206 the selector elements to be different from the data elements
5207 if the selector is a VECTOR_CST.
5208 * tree-vect-stmts.c (vect_gen_perm_mask_any): Use a vector of
5209 ssizetype for the selector.
5211 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5213 * optabs.c (shift_amt_for_vec_perm_mask): Try using series_p
5214 before testing each element individually.
5215 * tree-vect-generic.c (lower_vec_perm): Likewise.
5217 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5219 * selftest.h (selftest::vec_perm_indices_c_tests): Declare.
5220 * selftest-run-tests.c (selftest::run_tests): Call it.
5221 * vector-builder.h (vector_builder::operator ==): New function.
5222 (vector_builder::operator !=): Likewise.
5223 * vec-perm-indices.h (vec_perm_indices::series_p): Declare.
5224 (vec_perm_indices::all_from_input_p): New function.
5225 * vec-perm-indices.c (vec_perm_indices::series_p): Likewise.
5226 (test_vec_perm_12, selftest::vec_perm_indices_c_tests): Likewise.
5227 * fold-const.c (fold_ternary_loc): Use tree_to_vec_perm_builder
5228 instead of reading the VECTOR_CST directly. Detect whether both
5229 vector inputs are the same before constructing the vec_perm_indices,
5230 and update the number of inputs argument accordingly. Use the
5231 utility functions added above. Only construct sel2 if we need to.
5233 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5235 * optabs.c (expand_vec_perm_var): Use an explicit encoding for
5236 the broadcast of the low byte.
5237 (expand_mult_highpart): Use an explicit encoding for the permutes.
5238 * optabs-query.c (can_mult_highpart_p): Likewise.
5239 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Likewise.
5240 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5241 (vectorizable_bswap): Likewise.
5242 * tree-vect-data-refs.c (vect_grouped_store_supported): Use an
5243 explicit encoding for the power-of-2 permutes.
5244 (vect_permute_store_chain): Likewise.
5245 (vect_grouped_load_supported): Likewise.
5246 (vect_permute_load_chain): Likewise.
5248 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5250 * vec-perm-indices.h (vec_perm_indices_to_tree): Declare.
5251 * vec-perm-indices.c (vec_perm_indices_to_tree): New function.
5252 * tree-ssa-forwprop.c (simplify_vector_constructor): Use it.
5253 * tree-vect-slp.c (vect_transform_slp_perm_load): Likewise.
5254 * tree-vect-stmts.c (vectorizable_bswap): Likewise.
5255 (vect_gen_perm_mask_any): Likewise.
5257 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5259 * int-vector-builder.h: New file.
5260 * vec-perm-indices.h: Include int-vector-builder.h.
5261 (vec_perm_indices): Redefine as an int_vector_builder.
5262 (auto_vec_perm_indices): Delete.
5263 (vec_perm_builder): Redefine as a stand-alone class.
5264 (vec_perm_indices::vec_perm_indices): New function.
5265 (vec_perm_indices::clamp): Likewise.
5266 * vec-perm-indices.c: Include fold-const.h and tree-vector-builder.h.
5267 (vec_perm_indices::new_vector): New function.
5268 (vec_perm_indices::new_expanded_vector): Update for new
5269 vec_perm_indices class.
5270 (vec_perm_indices::rotate_inputs): New function.
5271 (vec_perm_indices::all_in_range_p): Operate directly on the
5272 encoded form, without computing elided elements.
5273 (tree_to_vec_perm_builder): Operate directly on the VECTOR_CST
5274 encoding. Update for new vec_perm_indices class.
5275 * optabs.c (expand_vec_perm_const): Create a vec_perm_indices for
5276 the given vec_perm_builder.
5277 (expand_vec_perm_var): Update vec_perm_builder constructor.
5278 (expand_mult_highpart): Use vec_perm_builder instead of
5279 auto_vec_perm_indices.
5280 * optabs-query.c (can_mult_highpart_p): Use vec_perm_builder and
5281 vec_perm_indices instead of auto_vec_perm_indices. Use a single
5282 or double series encoding as appropriate.
5283 * fold-const.c (fold_ternary_loc): Use vec_perm_builder and
5284 vec_perm_indices instead of auto_vec_perm_indices.
5285 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5286 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5287 (vect_permute_store_chain): Likewise.
5288 (vect_grouped_load_supported): Likewise.
5289 (vect_permute_load_chain): Likewise.
5290 (vect_shift_permute_load_chain): Likewise.
5291 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5292 (vect_transform_slp_perm_load): Likewise.
5293 (vect_schedule_slp_instance): Likewise.
5294 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5295 (vectorizable_mask_load_store): Likewise.
5296 (vectorizable_bswap): Likewise.
5297 (vectorizable_store): Likewise.
5298 (vectorizable_load): Likewise.
5299 * tree-vect-generic.c (lower_vec_perm): Use vec_perm_builder and
5300 vec_perm_indices instead of auto_vec_perm_indices. Use
5301 tree_to_vec_perm_builder to read the vector from a tree.
5302 * tree-vect-loop.c (calc_vec_perm_mask_for_shift): Take a
5303 vec_perm_builder instead of a vec_perm_indices.
5304 (have_whole_vector_shift): Use vec_perm_builder and
5305 vec_perm_indices instead of auto_vec_perm_indices. Leave the
5306 truncation to calc_vec_perm_mask_for_shift.
5307 (vect_create_epilog_for_reduction): Likewise.
5308 * config/aarch64/aarch64.c (expand_vec_perm_d::perm): Change
5309 from auto_vec_perm_indices to vec_perm_indices.
5310 (aarch64_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5311 instead of changing individual elements.
5312 (aarch64_vectorize_vec_perm_const): Use new_vector to install
5313 the vector in d.perm.
5314 * config/arm/arm.c (expand_vec_perm_d::perm): Change
5315 from auto_vec_perm_indices to vec_perm_indices.
5316 (arm_expand_vec_perm_const_1): Use rotate_inputs on d.perm
5317 instead of changing individual elements.
5318 (arm_vectorize_vec_perm_const): Use new_vector to install
5319 the vector in d.perm.
5320 * config/powerpcspe/powerpcspe.c (rs6000_expand_extract_even):
5321 Update vec_perm_builder constructor.
5322 (rs6000_expand_interleave): Likewise.
5323 * config/rs6000/rs6000.c (rs6000_expand_extract_even): Likewise.
5324 (rs6000_expand_interleave): Likewise.
5326 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5328 * optabs-query.c (can_vec_perm_var_p): Check whether lowering
5329 to qimode could truncate the indices.
5330 * optabs.c (expand_vec_perm_var): Likewise.
5332 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5334 * Makefile.in (OBJS): Add vec-perm-indices.o.
5335 * vec-perm-indices.h: New file.
5336 * vec-perm-indices.c: Likewise.
5337 * target.h (vec_perm_indices): Replace with a forward class
5339 (auto_vec_perm_indices): Move to vec-perm-indices.h.
5340 * optabs.h: Include vec-perm-indices.h.
5341 (expand_vec_perm): Delete.
5342 (selector_fits_mode_p, expand_vec_perm_var): Declare.
5343 (expand_vec_perm_const): Declare.
5344 * target.def (vec_perm_const_ok): Replace with...
5345 (vec_perm_const): ...this new hook.
5346 * doc/tm.texi.in (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Replace with...
5347 (TARGET_VECTORIZE_VEC_PERM_CONST): ...this new hook.
5348 * doc/tm.texi: Regenerate.
5349 * optabs.def (vec_perm_const): Delete.
5350 * doc/md.texi (vec_perm_const): Likewise.
5351 (vec_perm): Refer to TARGET_VECTORIZE_VEC_PERM_CONST.
5352 * expr.c (expand_expr_real_2): Use expand_vec_perm_const rather than
5353 expand_vec_perm for constant permutation vectors. Assert that
5354 the mode of variable permutation vectors is the integer equivalent
5355 of the mode that is being permuted.
5356 * optabs-query.h (selector_fits_mode_p): Declare.
5357 * optabs-query.c: Include vec-perm-indices.h.
5358 (selector_fits_mode_p): New function.
5359 (can_vec_perm_const_p): Check whether targetm.vectorize.vec_perm_const
5360 is defined, instead of checking whether the vec_perm_const_optab
5361 exists. Use targetm.vectorize.vec_perm_const instead of
5362 targetm.vectorize.vec_perm_const_ok. Check whether the indices
5363 fit in the vector mode before using a variable permute.
5364 * optabs.c (shift_amt_for_vec_perm_mask): Take a mode and a
5365 vec_perm_indices instead of an rtx.
5366 (expand_vec_perm): Replace with...
5367 (expand_vec_perm_const): ...this new function. Take the selector
5368 as a vec_perm_indices rather than an rtx. Also take the mode of
5369 the selector. Update call to shift_amt_for_vec_perm_mask.
5370 Use targetm.vectorize.vec_perm_const instead of vec_perm_const_optab.
5371 Use vec_perm_indices::new_expanded_vector to expand the original
5372 selector into bytes. Check whether the indices fit in the vector
5373 mode before using a variable permute.
5374 (expand_vec_perm_var): Make global.
5375 (expand_mult_highpart): Use expand_vec_perm_const.
5376 * fold-const.c: Includes vec-perm-indices.h.
5377 * tree-ssa-forwprop.c: Likewise.
5378 * tree-vect-data-refs.c: Likewise.
5379 * tree-vect-generic.c: Likewise.
5380 * tree-vect-loop.c: Likewise.
5381 * tree-vect-slp.c: Likewise.
5382 * tree-vect-stmts.c: Likewise.
5383 * config/aarch64/aarch64-protos.h (aarch64_expand_vec_perm_const):
5385 * config/aarch64/aarch64-simd.md (vec_perm_const<mode>): Delete.
5386 * config/aarch64/aarch64.c (aarch64_expand_vec_perm_const)
5387 (aarch64_vectorize_vec_perm_const_ok): Fuse into...
5388 (aarch64_vectorize_vec_perm_const): ...this new function.
5389 (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5390 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5391 * config/arm/arm-protos.h (arm_expand_vec_perm_const): Delete.
5392 * config/arm/vec-common.md (vec_perm_const<mode>): Delete.
5393 * config/arm/arm.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5394 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5395 (arm_expand_vec_perm_const, arm_vectorize_vec_perm_const_ok): Merge
5397 (arm_vectorize_vec_perm_const): ...this new function. Explicitly
5398 check for NEON modes.
5399 * config/i386/i386-protos.h (ix86_expand_vec_perm_const): Delete.
5400 * config/i386/sse.md (VEC_PERM_CONST, vec_perm_const<mode>): Delete.
5401 * config/i386/i386.c (ix86_expand_vec_perm_const_1): Update comment.
5402 (ix86_expand_vec_perm_const, ix86_vectorize_vec_perm_const_ok): Merge
5404 (ix86_vectorize_vec_perm_const): ...this new function. Incorporate
5405 the old VEC_PERM_CONST conditions.
5406 * config/ia64/ia64-protos.h (ia64_expand_vec_perm_const): Delete.
5407 * config/ia64/vect.md (vec_perm_const<mode>): Delete.
5408 * config/ia64/ia64.c (ia64_expand_vec_perm_const)
5409 (ia64_vectorize_vec_perm_const_ok): Merge into...
5410 (ia64_vectorize_vec_perm_const): ...this new function.
5411 * config/mips/loongson.md (vec_perm_const<mode>): Delete.
5412 * config/mips/mips-msa.md (vec_perm_const<mode>): Delete.
5413 * config/mips/mips-ps-3d.md (vec_perm_constv2sf): Delete.
5414 * config/mips/mips-protos.h (mips_expand_vec_perm_const): Delete.
5415 * config/mips/mips.c (mips_expand_vec_perm_const)
5416 (mips_vectorize_vec_perm_const_ok): Merge into...
5417 (mips_vectorize_vec_perm_const): ...this new function.
5418 * config/powerpcspe/altivec.md (vec_perm_constv16qi): Delete.
5419 * config/powerpcspe/paired.md (vec_perm_constv2sf): Delete.
5420 * config/powerpcspe/spe.md (vec_perm_constv2si): Delete.
5421 * config/powerpcspe/vsx.md (vec_perm_const<mode>): Delete.
5422 * config/powerpcspe/powerpcspe-protos.h (altivec_expand_vec_perm_const)
5423 (rs6000_expand_vec_perm_const): Delete.
5424 * config/powerpcspe/powerpcspe.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK):
5426 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5427 (altivec_expand_vec_perm_const_le): Take each operand individually.
5428 Operate on constant selectors rather than rtxes.
5429 (altivec_expand_vec_perm_const): Likewise. Update call to
5430 altivec_expand_vec_perm_const_le.
5431 (rs6000_expand_vec_perm_const): Delete.
5432 (rs6000_vectorize_vec_perm_const_ok): Delete.
5433 (rs6000_vectorize_vec_perm_const): New function.
5434 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5435 an element count and rtx array.
5436 (rs6000_expand_extract_even): Update call accordingly.
5437 (rs6000_expand_interleave): Likewise.
5438 * config/rs6000/altivec.md (vec_perm_constv16qi): Delete.
5439 * config/rs6000/paired.md (vec_perm_constv2sf): Delete.
5440 * config/rs6000/vsx.md (vec_perm_const<mode>): Delete.
5441 * config/rs6000/rs6000-protos.h (altivec_expand_vec_perm_const)
5442 (rs6000_expand_vec_perm_const): Delete.
5443 * config/rs6000/rs6000.c (TARGET_VECTORIZE_VEC_PERM_CONST_OK): Delete.
5444 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5445 (altivec_expand_vec_perm_const_le): Take each operand individually.
5446 Operate on constant selectors rather than rtxes.
5447 (altivec_expand_vec_perm_const): Likewise. Update call to
5448 altivec_expand_vec_perm_const_le.
5449 (rs6000_expand_vec_perm_const): Delete.
5450 (rs6000_vectorize_vec_perm_const_ok): Delete.
5451 (rs6000_vectorize_vec_perm_const): New function. Remove stray
5452 reference to the SPE evmerge intructions.
5453 (rs6000_do_expand_vec_perm): Take a vec_perm_builder instead of
5454 an element count and rtx array.
5455 (rs6000_expand_extract_even): Update call accordingly.
5456 (rs6000_expand_interleave): Likewise.
5457 * config/sparc/sparc.md (vec_perm_constv8qi): Delete in favor of...
5458 * config/sparc/sparc.c (sparc_vectorize_vec_perm_const): ...this
5460 (TARGET_VECTORIZE_VEC_PERM_CONST): Redefine.
5462 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5464 * optabs.c (expand_vec_perm_1): Assert that SEL has an integer
5465 vector mode and that that mode matches the mode of the data
5467 (expand_vec_perm): Split handling of non-CONST_VECTOR selectors
5468 out into expand_vec_perm_var. Do all CONST_VECTOR handling here,
5469 directly using expand_vec_perm_1 when forcing selectors into
5471 (expand_vec_perm_var): New function, split out from expand_vec_perm.
5473 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5475 * optabs-query.h (can_vec_perm_p): Delete.
5476 (can_vec_perm_var_p, can_vec_perm_const_p): Declare.
5477 * optabs-query.c (can_vec_perm_p): Split into...
5478 (can_vec_perm_var_p, can_vec_perm_const_p): ...these two functions.
5479 (can_mult_highpart_p): Use can_vec_perm_const_p to test whether a
5480 particular selector is valid.
5481 * tree-ssa-forwprop.c (simplify_vector_constructor): Likewise.
5482 * tree-vect-data-refs.c (vect_grouped_store_supported): Likewise.
5483 (vect_grouped_load_supported): Likewise.
5484 (vect_shift_permute_load_chain): Likewise.
5485 * tree-vect-slp.c (vect_build_slp_tree_1): Likewise.
5486 (vect_transform_slp_perm_load): Likewise.
5487 * tree-vect-stmts.c (perm_mask_for_reverse): Likewise.
5488 (vectorizable_bswap): Likewise.
5489 (vect_gen_perm_mask_checked): Likewise.
5490 * fold-const.c (fold_ternary_loc): Likewise. Don't take
5491 implementations of variable permutation vectors into account
5492 when deciding which selector to use.
5493 * tree-vect-loop.c (have_whole_vector_shift): Don't check whether
5494 vec_perm_const_optab is supported; instead use can_vec_perm_const_p
5495 with a false third argument.
5496 * tree-vect-generic.c (lower_vec_perm): Use can_vec_perm_const_p
5497 to test whether the constant selector is valid and can_vec_perm_var_p
5498 to test whether a variable selector is valid.
5500 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5502 * optabs-query.h (can_vec_perm_p): Take a const vec_perm_indices *.
5503 * optabs-query.c (can_vec_perm_p): Likewise.
5504 * fold-const.c (fold_vec_perm): Take a const vec_perm_indices &
5505 instead of vec_perm_indices.
5506 * tree-vectorizer.h (vect_gen_perm_mask_any): Likewise,
5507 (vect_gen_perm_mask_checked): Likewise,
5508 * tree-vect-stmts.c (vect_gen_perm_mask_any): Likewise,
5509 (vect_gen_perm_mask_checked): Likewise,
5511 2018-01-02 Richard Sandiford <richard.sandiford@linaro.org>
5513 * optabs-query.h (qimode_for_vec_perm): Declare.
5514 * optabs-query.c (can_vec_perm_p): Split out qimode search to...
5515 (qimode_for_vec_perm): ...this new function.
5516 * optabs.c (expand_vec_perm): Use qimode_for_vec_perm.
5518 2018-01-02 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
5520 * rtlanal.c (canonicalize_condition): Return 0 if final rtx
5521 does not have a conditional at the top.
5523 2018-01-02 Richard Biener <rguenther@suse.de>
5525 * ipa-inline.c (big_speedup_p): Fix expression.
5527 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5530 * config/i386/x86-tune-costs.h: Increase cost of integer load costs
5533 2018-01-02 Jan Hubicka <hubicka@ucw.cz>
5537 * x86-tune-costs.h (generic_cost): Reduce cost of FDIV 20->17,
5538 cost of sqrt 20->14, DIVSS 18->13, DIVSD 32->17, SQRtSS 30->14
5539 and SQRTsD 58->18, cond_not_taken_branch_cost. 2->1. Increase
5540 cond_taken_branch_cost 3->4.
5542 2018-01-01 Jakub Jelinek <jakub@redhat.com>
5544 PR tree-optimization/83581
5545 * tree-loop-distribution.c (pass_loop_distribution::execute): Return
5546 TODO_cleanup_cfg if any changes have been made.
5549 * expr.c (store_expr_with_bounds): Use simplify_gen_subreg instead of
5550 convert_modes if target mode has the right side, but different mode
5554 * expr.c (expand_assignment): Fix up a typo in simplify_gen_subreg
5555 last argument when extracting from CONCAT. If either from_real or
5556 from_imag is NULL, use expansion through memory. If result is not
5557 a CONCAT and simplify_gen_subreg fails, try to simplify_gen_subreg
5558 the parts directly to inner mode, if even that fails, use expansion
5562 * expmed.c (expand_shift_1): For 2-byte rotates by BITS_PER_UNIT,
5563 check for bswap in mode rather than HImode and use that in expand_unop
5566 Copyright (C) 2018 Free Software Foundation, Inc.
5568 Copying and distribution of this file, with or without modification,
5569 are permitted in any medium without royalty provided the copyright
5570 notice and this notice are preserved.